AI Video Stack 2026: Live Comparison & Rankings

This is RCTV’s living reference to the AI video software stack — models, orchestration agents, and open-source generation. Updated as products launch, pricing changes, and capabilities evolve. Last updated: October 11, 2026.

The short answer (October 2026). There is no single best AI video model — the right pick is set by the axis you optimize. For photorealism, Veo 3.1 (4K native, free on any Google account). For broadcast-grade 4K motion, Kling 3.0 (native 4K at 60fps; Kuaishou closed a record ~$3B round at an ~$18B valuation in July 2026, the clearest capital signal in AI video to date). For character consistency across shots, Seedance 2.0 Pro (ByteDance). For stylized and VFX work, Runway; for editing existing footage, Runway’s Aleph 2.0. For free consumer distribution, Gemini Omni Flash (Google; multimodal input, SynthID + C2PA provenance by default). For open-weights local generation, LTX-2.5 (true 4K on consumer GPUs; LTX-2.x Community License, free below $10M revenue, not OSI-open). For top benchmark quality, Wan 3.0 (Alibaba, invite-gated) leads Artificial Analysis’s rebuilt text-to-video boards with and without audio and, with audio, its video-editing board, and MiniMax H3 Max leads image-to-video with audio; Gemini Omni Flash’s July sweep of all four boards is over. Grok Imagine Video 1.5 lists at $0.08/sec at 480p, $0.14 at 720p and $0.25 at 1080p, with native audio. The larger 2026 shift: the models are converging, and the contest has moved off raw model quality to the orchestration and assembly layer built on top of them (see The Agentic Layer).

Quick Verdict: Best AI Video Model by Use Case (2026)

Routing decisions based on what each model actually wins at — not benchmark Elo alone, but the practical question of which model handles a specific use case best. Verdicts current as of October 11, 2026. Scroll down for full model breakdowns and the detailed routing framework.

• Best for photorealism — Veo 3.1. 4K native, the cleanest photorealistic rendering available; free 10 clips/month via Google Vids on any Google account.
• Best for broadcast 4K and motion quality — Kling 3.0. Native 4K at 60fps with multi-cut storyboards; the only production-grade model meeting broadcast delivery standards without upscaling.
• Best for character consistency across shots — Seedance 2.0 Pro. Native multi-shot storytelling with frame-level character and scene control; available in the US via CapCut with real-face restrictions.
• Best for stylized and VFX work — Runway Gen-4 Turbo. The most mature professional ecosystem for non-photorealistic aesthetics; motion brushes, scene consistency tools, $315M Series C runway.
• Best for editing existing video — Aleph 2.0 in Runway Edit Studio. Multishot edit propagation up to 30 seconds at 1080p; edit one frame, model carries it across the sequence; available on all paid Runway plans.
• Best for free consumer distribution — Gemini Omni Flash. Free inside YouTube Shorts and YouTube Create; multimodal any-to-any input (image / audio / video / text → video); conversational multi-turn editing; SynthID + C2PA provenance by default.
• Best open-weights for local generation — LTX-2.5. True 4K native on consumer GPUs (12GB+ VRAM); standalone desktop editor plus ComfyUI integration. Open weights under the LTX-2.x Community License, not an OSI-open licence — free below $10M annual revenue, and it bars training a competing model on the outputs. On measured quality it trails: 23rd (Fast, Elo 946) and 24th (Pro, 943) of 29 on Artificial Analysis’s rebuilt text-to-video v2.0 board with audio as of October 11, 2026 — a tie inside the error bars, not a gap between tiers.
• Best benchmark quality — Wan 3.0 on text-to-video and, with audio, video-editing; MiniMax H3 Max on image-to-video with audio. On Artificial Analysis’s rebuilt text-to-video v2.0 board (September 30, every model judged at 1080p), Wan 3.0 leads with audio at 1,156 (rank range 1–2) over Utopai X (1,149, post-trained on H3, no public API), Dreamina Seedance 2.5 (1,143) and MiniMax H3 (1,137); on the silent board it leads at 1,129 with H3 Max (1,118) and H3 (1,117) next. Gemini Omni Flash 1.1 is eighth on both (1,113 / 1,069). Wan 3.0 also holds video-editing with audio (1,187) and H3 Max leads image-to-video with audio (1,195), on boards that are not on the v2.0 scale (video-editing is v1.1, image-to-video v1.0). No v2.0 number belongs beside a v1.x one.
• Resolution-tiered API price — Grok Imagine Video 1.5. $0.08/sec at 480p, $0.14 at 720p and $0.25 at 1080p ($15.00 a minute), per xAI’s model page, with native audio; GA as of June 16, 2026. Native 1080p shipped July 31, 2026, closing the resolution gap tracked since Musk’s missed April commitment. Its lower-cost sibling Grok Imagine Video 1.5 Lite lists at $0.02 / $0.03 / $0.14. Arena positions in the entry.
• Best multi-model orchestration agent — Pika Agents. Broadest model roster in the category (Kling, Veo, Seedance, MiniMax, Sora API, Pika Video); runs inside Slack, Telegram, Discord, Notion, Figma, and 12+ other surfaces; persistent memory across sessions. See The Agentic Layer.
• Best for generative video editing — Wan 3.0. #1 on Artificial Analysis’s video-editing leaderboard with audio (Elo 1,187, October 11, 2026), ahead of Dreamina Seedance 2.5 (1,151), Gemini Omni Flash 1.1 (1,139) and MiniMax H3 (1,119, fourth). On the No Audio toggle Omni Flash 1.1 is first (1,157) over Wan 3.0 (1,146), inside the error bars. Wan 3.0 took the category from H3, which had held it since its August 3 debut. Wan 3.0 is invite-gated commercial beta; Seedance 2.5 and H3 are the ones you can buy today at published API prices.
• Fastest generation — MiniMax H3 Max (post-trained by fal). A 5-second 768p clip in under 3 seconds — generation faster than playback, the threshold Weekly Roundup — August 31, 2026 covered. Free tier, no signup for the first tier. The sustained-load behaviour is unproven: fal’s own livestream product built on it was paused three hours after launch and was running six of nine channels a week later.

Quick Reference: All Models at a Glance

Rankings and pricing change weekly; arena positions are Artificial Analysis unless noted, captured October 11, 2026. Scroll down for full model breakdowns.

How to read the leaderboard figures. Artificial Analysis runs three scored leaderboard pages — text-to-video, image-to-video and video-editing — and each one carries a With Audio / No Audio toggle that swaps the entire ranking and every Elo value on the same URL. They are no longer on one scale. Text-to-video was rebuilt on September 30 as AA-Video-T2V v2.0 and T2V-Silent v2.0: every model judged at 1080p on a new prompt set, Kling 3.0 1080p (Pro) fixed at exactly 1,000. Every earlier text-to-video Elo is on a retired scale, and no old-scale figure belongs beside a v2.0 one. Image-to-video is AA-Video-I2V v1.0 (AA says v2.0 is coming, with 1080p throughout) and video-editing is AA-Video-Editing v1.1; each keeps its own scale, and an Elo means something only within its own board — video-editing is anchored on a different Kling entry. The separate text-to-video-with-audio address retired in August 2026. Figures on this page are the With Audio set unless stated otherwise. As of October 11, 2026: text-to-video with audio runs Wan 3.0 (1,156), Utopai X (1,149), Dreamina Seedance 2.5 (1,143), MiniMax H3 at 768p (1,137), H3 Max (1,130); silent runs Wan 3.0 (1,129), H3 Max (1,118), H3 (1,117), Seedance 2.5 (1,105), FLUX 3 (1,091). Image-to-video with audio runs H3 Max (1,195), H3 (1,181), Vidu Q4 Preview (1,179, provisional), Omni Flash (1,178), Seedance 2.0 720p (1,176), HiDream-O1-Video (1,175), with AA’s rank ranges putting ranks two through six in one overlapping group; no audio, Omni Flash (1,368), Bach 1.0 Pro (1,362), Wan 3.0 (1,362), H3 (1,356), Vimoo 1.0 (1,348). Video-editing with audio runs Wan 3.0 (1,187), Seedance 2.5 (1,151), Omni Flash 1.1 (1,139), H3 at 768p (1,119), HappyHorse-1.0 (1,088); no audio, Omni Flash 1.1 (1,157), Wan 3.0 (1,146), Seedance 2.5 (1,139), H3 (1,121), Seedance 2.0 (1,116). (Corrected 2026-09-20: this note said Omni Flash led no-audio text-to-video at 1,324; the no-audio toggle had not been re-read since it was written.) (Superseded 2026-10-11: this note called video-editing v1.0 and gave H3 second at 1,132; the page header reads v1.1 and Seedance 2.5 is second.)

The Big Ten: Commercial Models

These are the production-grade models dominating professional and creator workflows in 2026. The market has matured to the point where no single model leads across all dimensions — the professional standard is now multi-model routing, choosing the right tool for each specific shot.

Luma Ray 3.2 / Ray 3.14 — Luma AI

Best for: Professional production volume, frame-level keyframe control, HDR/EXR output, cost-efficient multi-shot workflows

• Max resolution: 1080p native
• Max duration: Up to 20 seconds at 1080p (Ray 3.2); variable (Ray 3.14)
• Key features (Ray 3.2, June 9, 2026): Up to 16 keyframes per clip for precise motion direction; performance tracking across up to 8 faces simultaneously; native HDR generation with 16-bit EXR export; Enhanced Reframe (aspect ratio, frame extension, background replacement); API launch (first API availability for Ray 3.2)
• Key features (Ray 3.14): Duration-change and loop workflows; Ray3 Modify (hybrid performance/acting control)
• Timeline editor / EDL Export (June 19, 2026): Multi-clip timeline editor with Edit Decision List export for NLE handoff; integrates with Ray 3.2 generation workflow
• Luma Connectors (June 24, 2026): Native integrations with Airtable, Dropbox, and Google Drive for asset management and workflow automation
• Speed: 4× faster generation than previous Ray model (Ray 3.14 baseline); Ray 3.2 maintains the same 1080p baseline
• Pricing: 3× cheaper per-second than previous Ray (Ray 3.14 baseline)
• Access: Luma AI subscription; free tier available; a named partner model in Adobe’s Premiere Pro / After Effects Generative Media Tool (September 8, 2026) — and absent from Runway’s plugin into the same apps (Weekly Roundup — September 14, 2026)
• API: Available for both variants; Ray 3.2 API launched June 9, 2026; enterprise deployments via Luma Agents
• Note: Ray 3.2 and Ray 3.14 are parallel sub-models in the Ray3 family. Ray 3.2 handles standard video-to-video transformation and multi-keyframe guidance; Ray 3.14 handles duration-change and loop workflows. Both remain available.

Luma AI’s Ray 3.14 shipped in January 2026 as the model that stepped into the commercial tier vacated by Sora’s shutdown. (Weekly Roundup — March 27, 2026) Native 1080p output, generation 4× faster than the previous Ray 3 model, per-second pricing 3× cheaper. Ray3 Modify, a companion tool for hybrid performance and acting workflows, gives studios more control over scene continuity and character consistency across shots.

Luma launched Ray 3.2 on June 9, 2026 — framing the update as a shift “from prompting to directing.” The headline addition is keyframe control: up to 16 keyframes per clip at arbitrary positions, letting operators set precise motion direction and pacing rather than prompting for results. Alongside it: performance tracking across up to 8 simultaneous faces, native HDR generation with 16-bit EXR export for post-production pipelines, Enhanced Reframe for aspect-ratio changes and background replacement, and full API availability — the first Ray 3.2 API access. Duration extends to 20 seconds at 1080p. Ray 3.14 stays available and is the better choice specifically for duration-change and loop workflows, where its fixed-length output is an asset rather than a constraint.

Luma is positioning Ray explicitly as professional infrastructure priced for production volume rather than a consumer app — a distinction that looks strategically deliberate given Sora’s failure. The company’s $900M Series C led by HUMAIN, a London office, and enterprise Luma Agents deployments at Publicis, Adidas, and Mazda all reinforce this direction. The Mazda relationship produced a concrete deliverable in April 2026: Boundless, a Johannesburg agency, used Luma Agents to deliver Mazda’s first AI-produced commercial in under two weeks — the most credible production-deployment signal for any AI video platform to date.

Kling 3.0 / Kling 3.0 Omni / Kling 3.0 Turbo — Kuaishou

Best for: Feature density, broadcast-ready output, motion quality; Turbo for rapid iteration and cost-efficiency

• Max resolution: 4K native (60fps) — Standard and O3; 720p and 1080p on the API price list — Turbo
• Frame rate: Up to 60fps
• Audio: Native built-in audio in six languages
• Key feature: Multi-cut storyboard generation (up to 6 camera cuts, 15s); Omni/O3 adds shot/camera/character controls; Turbo adds faster generation, lower cost, improved lip-sync, stable motion
• Kling 3.0 Turbo (June 17, 2026): Speed-and-cost tier; faster generation, lower price, improved lip-sync, stable motion; targets rapid iteration workflows
• Kling 3.0 O3 (June 17, 2026): Up to 15-second clips at full 4K; stronger prompt-and-reference consistency; targets production-quality delivery
• Kling MCP & CLI (July 8, 2026): Kling’s model is now agent-callable — a Model Context Protocol endpoint plus a command-line interface, joining the MCP-exposed tier alongside Runway, Pika, and Higgsfield (Weekly Roundup — July 13, 2026)
• Platform rollout (June 17–18, 2026): fal, SeaArt, Clipfly, Fotor, GlamAI, Runware, Morphic — seven partner platforms in a coordinated same-day launch
• Pricing (API, per kling.ai/dev/pricing, read October 11, 2026; 1 unit = $0.14 list): Kling 3.0 Turbo with native audio $0.112/sec at 720p and $0.14/sec at 1080p; Kling 3.0 $0.084 / $0.112 / $0.42 per second at 720p / 1080p / 4K without native audio, $0.126 / $0.168 / $0.42 with it; Kling 3.0 Omni $0.084 / $0.112 / $0.42 without video input or native audio, $0.112 / $0.14 / $0.42 with native audio. No Kling 4.0 or Flash row.
• Access: Free tier available; paid plans from ~$8/mo; also via Adobe Firefly (Creative Cloud subscription); from September 8, 2026 a named partner model in Adobe’s Premiere Pro / After Effects Generative Media Tool and, as Kling 3.0 Pro, in Runway’s plugin into the same apps (Weekly Roundup — September 14, 2026)
• API: Available via Kuaishou and third-party platforms
• Next: Kling 4.0 — announced, full launch promised for October, not live as of October 11, 2026; Flash is in limited early access. See What’s Coming.

The most capability-dense model available. Kling 3.0 is the first AI video model to meet broadcast delivery standards without upscaling, offering native 4K at 60fps. The storyboard feature generates up to six camera cuts in a single generation with visual consistency — a production-first capability no other model matches. The Kling 3.0 Omni (O3) variant adds finer-grained controls for shot duration, camera angle, and character movement across multi-shot sequences, with clips up to 15 seconds at full 4K.

On June 17–18, 2026, Kuaishou rolled two new Kling variants to seven partner platforms simultaneously — fal, SeaArt, Clipfly, Fotor, GlamAI, Runware, and Morphic — with no direct Kuaishou announcement on klingai.com (which remains 446-blocked). Kling 3.0 Turbo targets speed and cost: faster generation, lower price, improved lip-sync, and more stable motion than Standard, designed for rapid iteration and volume workflows. Kling 3.0 O3 targets production quality: clips up to 15 seconds at full 4K, with stronger prompt and reference consistency. One tier for iteration; one for delivery. Kuaishou’s API pricing page lists both variants (see the pricing bullet above); the June 17–18 capability claims were sourced from partner-platform announcements, which agreed across all seven independently.

In April 2026, both Kling 3.0 and Kling 3.0 Omni joined Adobe Firefly’s multi-model video hub alongside Veo 3.1, Runway Gen-4.5, and 30+ other AI models — significantly broadening Kling’s distribution to Adobe Creative Cloud’s existing professional user base.

In July 2026 the financing closed. Kuaishou disclosed on July 2 a round of nearly $3 billion at an ~$18 billion post-money valuation — reported as the largest ever raised by an AI-video-model company. It was led by CPE Yuanfeng, Guofang Venture Capital, Tencent, Zhongguancun Science City Fund, and CITIC Securities, with Alibaba Cloud and Baidu among 38-plus investors; Kuaishou’s stake dilutes from 100% to 68.33% (outside investors plus an employee ownership plan), and the round is widely read as pre-positioning for a Hong Kong IPO within 12 months. The revenue underneath it is real: Q1 2026 revenue topped 650M yuan (about $96M), +300% YoY, an annualized run rate near $500M — quadruple a year earlier. It priced below the roughly $20B target floated in May, but stands as the clearest commercial-scale signal in AI video to date, and the first pure-play comparable since Sora’s shutdown removed one (Weekly Roundup — July 13, 2026). The scale behind it is now disclosed: Kling has crossed 100 million cumulative users across 224 countries, ARR ~$500M as of March 2026 (up from ~$240M in December 2025; ~70–75% overseas). At Cannes Lions 2026 — whose inaugural “AI Craft” Grand Prix went to Google, not Kling — two Kling-made ad films still won real Lions (Silver + Bronze for The RealReal’s “L’Ultimo Uomo Reale”) (Weekly Roundup — July 20, 2026).

Two more investors followed in August. Per Kuaishou’s stock-exchange announcement of August 31, 2026, as reported by TechNode, the state-backed China Artificial Intelligence Industry Investment Fund subscribed RMB1.4 billion for about 1.14% of Beijing Kling Technology, and Charoen Pokphand’s CP Robot about $19.29 million (RMB131.45 million) for about 0.11%, both with redemption rights. Small stakes at the same ~$18 billion mark as the July round; the name that matters is the fund’s — the Ministry of Finance is its largest ultimate shareholder, which puts state capital on Kling’s register ahead of the IPO the July investors are positioned for.

Veo 3.1 — Google DeepMind

Best for: Photorealism, 4K native output, integrated workflows, broadest free access

• Max resolution: 4K native (Flow/Vertex AI); 1080p via Veo 3.1 Lite; 720p via Google Vids free tier
• Audio: Native synchronized audio
• Key features: Flow unified workspace; Google Vids integration (avatars, Lyria 3 music, YouTube export); Veo 3.1 Lite developer tier; voice-driven generation on Gemini-enabled Google TV; Performance Control (July 2026): extracts precise 3D geometry and motion from a real performer to drive generation, with independently editable player/environment layers; debuted in Google DeepMind’s Pelé “Gol da Rua Javari” reconstruction
• Access: Free — 10 clips/month via Google Vids (any Google account); Google AI Pro ($19.99/mo) and Ultra for higher limits; Flow is free; also via Adobe Firefly multi-model hub (April 2026), Adobe’s Premiere Pro / After Effects Generative Media Tool and Runway’s plugin into the same apps (both September 8, 2026, (Weekly Roundup — September 14, 2026)) and Gemini-enabled TCL Google TVs in the US (April 2026)
• API: Vertex AI ($12/min); Veo 3.1 Lite via Gemini API ($0.05/sec 720p, $0.08/sec 1080p); Veo 3.1 Fast pricing reduced April 7, 2026 (check Gemini API docs for current per-second rates); the Veo 3.1, 3.1 Fast and 3.1 Lite preview model IDs shut down October 22, 2026, with gemini-omni-1.1-flash named as the replacement, per Google’s deprecations page
• Milestone: 1.5 billion images and videos created by Flow users

Google’s model pushes photorealistic rendering to the point where trained observers struggle to identify generated footage in blind tests. It is the engine behind Google Flow (merged creative workspace with Whisk, ImageFX, and multi-clip sequencing) and Google Vids. Veo 3.1 has been freely available to any Google account holder since April 2026 (Weekly Roundup — April 4, 2026) — 10 generations per month, 8 seconds at 720p, from text prompts or uploaded images. Google AI Pro and Ultra subscribers get more: up to 1,000 Veo clips per month, Lyria 3 custom music generation (tracks up to 3 minutes), customizable AI avatars with scene placement and wardrobe control, and direct YouTube export. This is the first time a production-grade AI video model has been made freely accessible to Google’s full account base.

Veo arrived on Gemini-enabled TCL Google TVs in the US in April 2026 (Weekly Roundup — May 11, 2026) — voice-driven generation through Gemini’s Create tab, either from scratch or by animating still images. TCL-only and US-only at launch, with no public timeline for other manufacturers or markets.

On the developer side, Google launched Veo 3.1 Lite in March 2026 via the Gemini API (Weekly Roundup — April 4, 2026) and Google AI Studio — priced at $0.05/sec for 720p and $0.08/sec for 1080p, less than half the cost of the existing Veo 3.1 Fast tier. Veo 3.1 Fast received a further price reduction on April 7, 2026 — compressing the full developer stack from the free consumer tier through production-grade API calls. Check the Gemini API pricing documentation for current per-second rates.

In July 2026, Google DeepMind showcased a new Veo technique, Performance Control, in its reconstruction of Pelé’s never-filmed 1959 goal — extracting 3D geometry and motion from a real stunt player to drive generation, with editable layers that swap the performer and the environment independently. The project pairs the technique with a live-action shoot and traditional VFX, and Google framed the output explicitly as a reconstruction, not archival footage — drawing its own AI-disclosure line ahead of the August 2 EU/California transparency deadlines (Weekly Roundup — July 27, 2026).

The technique’s third public showcase followed on September 9, 2026: “Love, Rendered”, a short documentary made with Primordial Soup, Darren Aronofsky’s production venture, that reconstructs the youth of a couple married 70 years from restored photographs, with performance-capture and pose-control models mapping their present-day mannerisms onto the reconstruction and one of the subjects correcting details as a co-creator. Google’s post is dated September 9; @GoogleDeepMind posted it September 11. Same capability class as the Pelé piece, applied to a private memory rather than a public archive.

Gemini Omni Flash — Google DeepMind

Best for: Multimodal input across image / audio / video / text → video output; conversational multi-turn editing; broadest free consumer distribution via YouTube Shorts; SynthID + C2PA provenance by default

• Max resolution: 720p is the only resolution Omni natively renders. Omni 1.1’s resolution parameter offers 360p/720p/1080p/4K, but Google’s own resolution table labels the top two “1080p output (upscaled)” and “4K output (upscaled)” in those words. A 4K Omni clip is a 720p render enlarged, not a 4K generation — the distinction that matters against Veo 3.1 and Kling 3.0, which render 4K natively
• Max duration: 10 seconds per generation (cap); “longer durations coming soon” per Google
• Audio: Native multimodal output — synchronized audio generated alongside video
• Watermark: SynthID watermarking plus C2PA Content Credentials embedded by default. Checkable through Google’s built-in verification in Search, the Gemini app and Chrome, and since October 7, 2026 in the standalone SynthID Detector, which takes images, audio and video, detects only SynthID marks from Google and adopting partners (OpenAI, NVIDIA and Kakao named, Apple listed as coming; per Google DeepMind), and says it is ’not a general AI detector’. Its terms bar bypassing the interface to submit requests programmatically. The Gemini app’s own check reads Google-made content only (Gemini Apps Help).
• Key feature: Any-to-any multimodal input (text / image / audio / video → video); conversational multi-turn editing via a stateful Interactions API that carries context across turns; reference-driven object and brand insertion
• Omni 1.1 Flash — GA August 27, 2026 as model ID gemini-omni-1.1-flash; gemini-omni-flash-preview was scheduled for deprecation on September 30, 2026; Google’s deprecations page, read October 4, no longer lists it and shows only gemini-omni-1.1-flash with no shutdown date. Adds scene extension, first- and last-frame specification, up to three video input references, 1080p/4K upscale, and a 360p fast mode for iteration. Scene extension works in 10-second increments to a cumulative 40-second ceiling, each extension using the last 10 seconds as context — so a 40-second Omni clip is four chained requests, not a single pass, and individual generations stay at 3–10 seconds. Read the distribution precisely: Google frames the full set as “for developers” and “available via APIs”; the Gemini app account scopes the consumer rollout to scene extension only — “available to all Google AI Plus, Pro and Ultra subscribers globally.” Five capabilities to developers; one of them to subscribers
• Access: Generally available as Omni 1.1 Flash since August 27, 2026 (public preview from June 30) via Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform; also live in the Gemini app and Google Flow; free via YouTube Shorts and YouTube Create
• API: $17.50 per 1M video output tokens at 5,792 tokens per second of 720p — about $0.10/sec, $6.00/min; input $1.50/1M. Same structure on the GA and preview rows
• Arena: Debuted #1 on Design Arena’s Video Arena at Elo 1,404 (July 2, 2026) — a 101-point margin over second place. Swept all four Artificial Analysis boards by mid-July 2026. That sweep is over. On the rebuilt text-to-video v2.0 board (September 30; a new scale, so no earlier text-to-video Elo compares) the board now scores Gemini Omni Flash 1.1 by name: #8 with audio (1,113) and #8 silent (1,069). On the boards still on v1.x scales it holds #4 image-to-video (1,178, a rank range of 2–6) and #3 video-editing with audio (1,139), is #1 on no-audio image-to-video (1,368) and #1 on no-audio video-editing (1,157, over Wan 3.0’s 1,146). The video-editing entry is Omni Flash 1.1 by name; the image-to-video entry is still the unversioned ‘Gemini Omni Flash’ (board release date May 19), so it describes the original model, not 1.1
• Next: Omni Pro shipping window (still no date); Artificial Analysis benchmark placement; durations past the 10-second preview cap

Google’s any-to-any multimodal video model entered public preview June 30, 2026, via Google AI Studio and the Gemini API — not the general availability that circulated in some coverage that week. Google’s own announcement is explicit: “Gemini Omni is available in public preview starting today.” The spec that shipped: $0.10 per second, 720p native with no 1080p or 4K tier — independently corroborated by VentureBeat’s pricing breakdown — clips capped at 10 seconds (“currently, with longer durations coming soon”), prompted in natural language and edited through a stateful Interactions API that carries context across turns: relight a shot, reframe it, change the wardrobe, without regenerating from scratch. Every output carries SynthID watermarking and C2PA Content Credentials by default.

This answers the question Google’s own I/O 2026 keynote left open: what does “a first step toward a world model” actually ship as? A price, a hard limit, and an admission — in Google’s own words — that “character consistency when changing scenes or panning movements has some limitations.”

Two days after the preview shipped, the first independent scoreboard weighed in: Omni Flash debuted #1 on Design Arena’s Video Arena at Elo 1,404 (July 2, 2026) — a 101-point margin over second place, one of the board’s largest gaps on record. The caveat mattered at preview: Design Arena isn’t Artificial Analysis, the leaderboard RCTV tracks for the rest of the Big Ten. It resolved fast — by mid-July 2026, Omni Flash had swept all four Artificial Analysis boards (with-audio T2V 1,240 / I2V 1,204, no-audio T2V 1,327 / I2V 1,374), passing both Seedance and HappyHorse — weeks after R#12 reported the AA score as pending. On Design Arena’s own image-to-video board, Seedance 2.0 still leads. A clean board sweep on an early vote count and a 10-second preview cap are both true at once (Weekly Roundup — July 6, 2026).

The sweep held about six weeks. By August 28, 2026 Wan 3.0 had taken text-to-video and video-editing and a fal fine-tune had taken image-to-video, leaving Omni Flash at #2/#4/#3 on scores that barely moved — the field caught up rather than the model regressing. Google’s answer, Omni 1.1 Flash on August 27, was not a bigger number: it was scene extension, frame control, video references, and 4K upscale. Controls, not score (Weekly Roundup — August 31, 2026). (Every text-to-video figure in this entry before September 30 is on the retired v1 scale.)

Seedance 2.0 Pro — ByteDance

Best for: Character consistency, cinematic motion, multi-shot storytelling

• Max resolution: 4K native, 10-bit color depth (confirmed at Volcano Engine FORCE, June 23, 2026); standard platform tiers vary
• Audio: Native audio with lip-sync
• Key feature: Multi-shot storytelling, quad-modal input, frame-level precision
• Access: China via Jimeng/Dreamina; Africa, South America, Middle East, SE Asia, and US via CapCut/Dreamina Seedance 2.0; global via BigMotion ($35–$95/mo), LumeFlow AI, other third-party platforms; unlimited access via Higgsfield (exclusive BytePlus-powered unlimited tier, June 17, 2026)
• API: Official global API paused; available via third-party integrations (fal.ai, Higgsfield/BytePlus, others)
• US restrictions: Real-face image-to-video disabled; unauthorized IP generation blocked; invisible watermarks on all output
• Seedance 2.5 rollout completed (August 21, 2026): Dreamina brought Seedance 2.5 to US users, closing the regional rollout that began August 14 across Southeast Asia, the Middle East, Africa, Europe and South America. Five assembly-layer platforms — Higgsfield (Aug 14), Dreamina and Runway (Aug 17), Luma and Pika (Aug 17) — were on 1080p Seedance 2.5 within four days (Weekly Roundup — August 24, 2026)
• Seedance 2.0 mini (June 16, 2026): Lightweight tier launched via Dreamina at approximately $0.02/sec for emerging markets (Southeast Asia, Middle East, Africa, Europe, South America); US-excluded; roughly 7× below the standard 720p tier (Artificial Analysis lists the standard tier at ~$0.151/sec, but fal.ai’s live 720p-with-audio rate runs higher at ~$0.30/sec — treat the live gateway figure as current for commercial pricing)
• Higgsfield “Seedance Unlimited” (June 17, 2026): 30-day unlimited video generation on “Enhanced Seedance 2.0 Fast” (a purpose-built speed-optimized model from ByteDance’s BytePlus B2B cloud); 1-day, 7-day, and 30-day add-on tiers for Higgsfield subscribers; 480p–720p; Higgsfield is the exclusive non-ByteDance unlimited-Seedance surface globally per BytePlus commercial agreement
• Dreamina Seedance 2.0 on US infrastructure via fal (September 10, 2026): fal’s announcement opens the model to US companies hosted entirely on US infrastructure — a deployment-sovereignty answer for a Chinese-origin model on a US inference host, distinct from the CapCut consumer path. Seedance 2.5 also ships inside Runway’s Premiere/After Effects plugin (September 8) while Adobe’s own timeline tool names no ByteDance model (Weekly Roundup — September 14, 2026)
• Note: Benchmark position (October 11, 2026): #7 text-to-video on the v2.0 board for the Dreamina Seedance 2.0 entry (Elo 1,116 with audio, 1,080 silent; Seedance 2.5 is #3 at 1,143), #5 image-to-video (Elo 1,176, v1.0 scale; the Dreamina Seedance 2.0 720p entry), #6–7 video-editing with audio (1,076, v1.1 scale; the Dreamina Seedance 2.0 entry), and Seedance 2.5 is #2 on video-editing with audio (1,151) and #3 without (1,139). It held #1 image-to-video until MiniMax H3 Max took it in late August. Seedance 2.5’s BytePlus API opened July 16 (see What’s Coming). Copyright legislative battle remains a three-way standoff (Blackburn vs. White House vs. CLEAR Act)

The leading commercial model for character consistency and cinematic motion quality. Seedance 2.0 Pro’s Dual-Branch Diffusion Transformer generates audio and video simultaneously in a single pass. Its quad-modal input system accepts text, images, video, and audio in a single prompt. Multi-shot native storytelling and frame-level control over character appearance, object placement, and scene timing remain best-in-class for narrative work.

ByteDance’s official global API rollout was paused indefinitely in late February 2026 after the Motion Picture Association and major studios (Disney, Netflix, Paramount, Sony, Warner Bros.) issued cease-and-desist letters over copyright concerns. The “Face-to-Voice” feature was suspended in February 2026 after it was shown to clone voices from a single photo. Japan opened a separate inquiry over unauthorized anime character reproductions.

ByteDance relaunched the model in March 2026 as Dreamina Seedance 2.0 (Weekly Roundup — March 27, 2026) across markets in Africa, South America, the Middle East, and Southeast Asia. As of April 2026, Dreamina Seedance 2.0 is available in the US via CapCut (Weekly Roundup — April 11, 2026) — a significant reversal of the prior exclusion. The deployment comes with content restrictions: image-to-video generation from inputs containing real faces is disabled, and generation of unauthorized intellectual property is blocked. All output carries an invisible watermark for off-platform identification.

In June 2026, ByteDance expanded Seedance’s distribution through two parallel channels. On June 16, Dreamina Seedance 2.0 mini launched for emerging markets — Southeast Asia, the Middle East, Africa, Europe, and South America — at approximately $0.02/sec, roughly 7× below the standard 720p tier. The US is excluded. On June 17, Higgsfield launched “Seedance Unlimited” — 30-day unlimited access to an “Enhanced Seedance 2.0 Fast” model, delivered through an exclusive partnership with BytePlus (ByteDance’s global B2B cloud platform). Higgsfield is the only non-ByteDance surface with unlimited Seedance globally; the commercial infrastructure runs through BytePlus while Higgsfield owns the creator-facing product. Two channels, one underlying model family, different regulatory profiles — the structure gives ByteDance revenue from Western markets without operating a direct consumer product there.

At Volcano Engine FORCE on June 23, 2026, ByteDance confirmed that Seedance 2.0 now supports native 4K with 10-bit color depth — the underlying capability specification the current three-platform distribution wave runs on. At the same event, ByteDance previewed Seedance 2.5; see What’s Coming.

The copyright landscape around Seedance is a three-way Washington standoff: the White House’s National Policy Framework for AI (March 2026) stated that AI training on copyrighted works does not constitute infringement — the opposite of the Blackburn bill’s position. Separately, the bipartisan CLEAR Act (Schiff/Curtis) would require public disclosure of training data without resolving the fair use question either way.

Grok Imagine — SpaceXAI (formerly xAI)

Best for: Speed, low-cost audio-native API, rapid iteration, social media distribution

• Max resolution: Original Grok Imagine: 720p. Grok Imagine Video 1.5: native 1080p as of July 31, 2026 — closing the resolution gap tracked since Musk’s missed April commitment (a separate, higher-end “Grok Imagine Pro” tier was never named again after the original announcement; 1.5’s own 1080p update appears to be the resolution fix that commitment pointed to)
• Arena (Grok Imagine): #5 on Artificial Analysis T2V no-audio on the retired v1 scale (Elo 1,235); the original Grok Imagine is not scored on the v2.0 boards
• Arena (Grok Imagine Video 1.5): #9 on the Artificial Analysis image-to-video board (Elo 1,098, October 11, 2026; AA rank range 8–10) — it opened at #2 in June with roughly the same score; seven newer entrants have since scored above it; on text-to-video v2.0, grok-imagine-video-1.5 is #11 with audio (1,045) and 12th silent (1,036)
• Grok Imagine Video 1.5 Lite (listed August 2026): a lower-cost sibling with its own xAI model page: text and image input, $0.02 / $0.03 / $0.14 per second at 480p / 720p / 1080p, plus $0.01 per input image. Artificial Analysis scores it on text-to-video v2.0 at #17 with audio (993) and #17 silent (984), at $8.40 a minute on its table (the 1080p list rate), and it is not on the with-audio image-to-video board. Runway’s API changelog, which added it to Runway Dev on October 1, says ‘1080p is rendered at 720p and upscaled’ and that Lite takes no image or audio references; xAI’s page says neither.
• Max duration: 15 seconds in a single request (added July 2026, per user reports amplified by Musk — no SpaceXAI release note); up to 30 seconds via chained extensions
• Audio: Original Grok Imagine: synchronized audio; Video 1.5: audio-native (synchronized in the same generation pass)
• First-and-last-frame control and seamless loops (September 8, 2026): per @imagine, with “improved instruction following and higher video quality” — a feature drop, not a model version
• Key feature: Video extension from frame; dual generation modes (Quality + Speed); native video understanding (Grok 4.3 Beta); fastest iteration cycle in the industry; Video 1.5 Fast variant (~25s for 6-sec 720p)
• Access: X Premium / SuperGrok subscription required
• API pricing:
  • Grok Imagine (original): $0.05/sec at 480p, $0.07 at 720p — no native audio
  • Grok Imagine Video 1.5 (GA June 16, 2026): $0.08/sec at 480p, $0.14 at 720p, $0.25 at 1080p — audio-native; the $0.08 headline on xAI’s page is the 480p rate. (Corrected 2026-10-11: this page said Grok Imagine Video 1.5 had a single published rate and no resolution tier on any xAI docs page. xAI’s model page lists $0.08, $0.14 and $0.25 per second at 480p, 720p and 1080p, and an archive capture of July 18 already showed the same table; $0.08 is the 480p rate. A 1080p minute lists at $15.00, not $4.80, the figure Artificial Analysis’s table already carried.)
• Engine: Aurora autoregressive MoE model on 110,000 NVIDIA GB200 GPUs
• Next: 1080p shipped in Video 1.5 (Jul 31, 2026) — voice references and text-to-video added the same update; watch whether the 1080p rate ($0.25/sec) moves now that resolution is no longer the differentiator
• Caution: Faced regulatory scrutiny over content moderation (UK ICO, France, California AG); image editing now restricted to paid subscribers

xAI shipped four major updates between January and March 2026: API launch (January 28), Grok Imagine 1.0 with 720p video and audio (February 3), Grok 4.20 (February 17), and video extension (March 2). The “Extend from Frame” feature lets users chain clips by continuing from the final frame, enabling sequences up to 30 seconds while preserving lighting, motion, and character positioning.

In April 2026, xAI released Grok 4.3 Beta with native video understanding — letting Grok analyze video as a coherent temporal sequence rather than as isolated frames. The understanding capability is distinct from Grok Imagine’s generation pipeline, but the two now stack: Grok can both generate and reason about video within the same model family. No other major lab currently offers vertical integration of native generation, native understanding, and platform-scale distribution under a single subscription.

On June 3, 2026, xAI shipped Grok Imagine Video 1.5 as a public API preview (Weekly Roundup — June 8, 2026) — the company’s first native image-to-video model. Per xAI’s docs: image-to-video with audio output, $0.08 per second, model alias grok-imagine-video-1.5-2026-05-30. It opened at #2 on the Artificial Analysis I2V with-audio arena, behind Seedance 2.0. Secondary reports described a higher-resolution tier; xAI’s model page lists per-second rates by resolution.

On June 16, 2026, xAI moved Grok Imagine Video 1.5 to general availability across the Imagine API, grok.com, iOS, and Android simultaneously. The GA launch added a Video 1.5 Fast variant — a 6-second 720p clip in approximately 25 seconds, down from 40-plus seconds in preview. xAI’s docs list $0.08 / $0.14 / $0.25 per second by resolution. At GA, Grok Imagine Video 1.5 holds #2 on the Artificial Analysis I2V with-audio arena (Elo 1,113), behind Seedance 2.0 (1,194). xAI’s #1 claim at launch wasn’t borne out by the live board. The “$4.20/min” figure circulating refers to the original Grok Imagine’s $0.07/sec@720p rate (a different, older model with no native audio) — not to Video 1.5.

Grok Imagine Video 1.5’s API list price is $0.08/sec at 480p, $0.14 at 720p and $0.25 at 1080p, with native audio, and as of August 2026 it no longer sits behind a resolution ceiling. The 1080p tier Elon Musk telegraphed for late April 2026 missed its April window by more than three months, but Grok Imagine Video 1.5 shipped native 1080p on July 31, 2026, alongside voice references and text-to-video generation — closing the gap this page has tracked since the missed commitment. 1080p is a priced tier at $0.25/sec, about 3.1 times the 480p rate. The launch puts the two subjects of RCTV’s Grok vs. Seedance comparison at one and two on the I2V arena.

SpaceX acquired xAI in an all-stock merger on February 2, 2026, valuing the combined company at $1.25 trillion — CNBC called it “the biggest merger of all time.” On July 6, 2026, the renamed company confirmed the shift on X: “We are now @SpaceXAI.” Nothing below changed with it — Grok, Grok Imagine, and the pricing and model names above are unchanged, and Artificial Analysis already credits the model to SpaceXAI.

Runway Gen-4 Turbo / Aleph 2.0 — Runway

Best for: Stylized content, VFX aesthetics, professional ecosystem, real-time avatars, agentic production, multishot edit propagation

• Max resolution: 1080p (Gen-4 Turbo); 1080p multishot up to 30s (Aleph 2.0 in Edit Studio); 720p real-time (Characters)
• Audio: Supported
• Key features: Motion brushes, style control, API maturity (Gen-4 Turbo); Characters real-time avatar API (GWM-1); Runway Agent (conversational end-to-end production, May 13, 2026); Aleph 2.0 multishot edit propagation in Edit Studio (May 21, 2026)
• Access: From $12/mo (runwayml.com); Gen-4.5 also via Adobe Firefly (Creative Cloud subscription); Runway Agent and Edit Studio (Aleph 2.0) at runwayml.com on all paid plans; the free Premiere Pro / After Effects plugin runs on the credits of whichever Runway plan the user already has
• API: Most mature video generation API available; Characters API at dev.runwayml.com
• Next: GWM Worlds 2 and Solaris moving past request-form access; standalone Gen-4.5 launch on runwayml.com still pending; real-time video model research preview on Vera Rubin hardware (sub-100ms TTF); independent hands-on for Aleph 2.0 against Veo 3.1 and Gemini Omni editing capabilities
• Dev API, August 31 – September 3, 2026: HDR outputs can be delivered as scene-referred ACEScg OpenEXR sequences (ACES 1.3 and 2.0 variants) on /v1/video_to_hdr and Gen-4.5 HDR generations, per the API changelog — a compositing-pipeline delivery format, not a generation capability; and MiniMax H3 Max joined the third-party roster on September 3 (see H3 Max)
• Ruby (August 21, 2026): A separate model converting SDR video — uploaded or Runway-generated, up to 30 seconds — to 16-bit HDR, delivering ProRes and EXR sequences at 10/12-bit BT.2020 with PQ or HLG. Max plan and Enterprise tiers only. A post-production/colour-pipeline tool rather than a generation capability, which is why it changes no benchmark position
• Solaris and GWM Worlds 2 (August 31 / September 3, 2026): two research previews from one world-model lineage, both early-access by request form. Solaris is an “Interface World Model” — it generates interactive software interfaces frame by frame (720p target, sub-500ms per frame), not video; Runway’s own 250-participant study preferred it to a Claude-Opus-5-coded interface 61–71% of the time depending on the comparison. GWM Worlds 2 is the video one: continuous 720p at 24fps with 48kHz audio, text-action steering addressed to any subject or the scene, no preset session length, successor to the GWM-1 line behind Characters. Neither is publicly available and neither has an independent evaluation (Weekly Roundup — September 7, 2026)
• Premiere Pro / After Effects plugin (September 8, 2026): a free, separately installed panel (Window → Extensions → Runway) that generates at the playhead — Gen-4.5, Seedance 2.5, Kling 3.0 Pro and Veo 3.1 for video, Aleph 2.0 to re-render a timeline clip, Ruby for HDR — billed on the user’s existing Runway credits. Adobe shipped its own native Generative Media Tool into the same two apps the same day with a different partner list (Firefly, Veo, Kling, Runway, Luma); neither company explains why Seedance is on one list and Luma on the other (Weekly Roundup — September 14, 2026)
• Runway Dev MCP (September 2, 2026): a hosted Model Context Protocol server at dev.runwayml.com/mcp that lets coding agents in Claude, ChatGPT, Codex and Cursor plan, configure and debug Runway Dev integrations, per Runway’s post — the developer-side counterpart to the May Runway MCP, no pricing stated
• Model licensing (September 11, 2026 post): @runwayml posted that enterprises can license Runway model weights to fine-tune on their own data, self-host and commercialize; Runway’s model-licensing page describes the package (weights, checkpoints, “white glove” support from the researchers who built the model; a World Action Model for robotics) and carries no date, so whether this is a fresh offer or a repromotion of the August 20 enterprise post is unresolved. Recorded as dated to the post, not as a launch
• Fall 2026 collection (September 16, 2026): Runway’s “biggest model drop yet” adds five models, none of them Runway’s — MiniMax H3 Max (video), Flux Video Upscale and Flux Video Edit (Black Forest Labs), Fish Audio S2.1 Pro and Cartesia Sonic 3.6 (audio). X-only; runway.com/news carries no matching post (Weekly Roundup — September 21, 2026)
• Moderation in Real Time (September 16, 2026): a synchronous content-moderation layer that screens generations before they render, built on Zentropi’s CoPE-B, a LoRA adapter over Google’s Gemma-4-26B-A4B-it; Runway’s own figure is an average review under 0.5 seconds, per the post, and the stated purpose is readiness for the more powerful real-time models Runway says it will release in the coming months (Weekly Roundup — September 21, 2026)
• Enhance Frame Rate (September 17, 2026): Runway’s own frame-interpolation model, per the post — output at 25, 30, 48, 60 or 120 fps including NTSC 59.94, one credit per two seconds of video regardless of resolution or target rate. A post-production utility, not a generation capability
• Unified pricing (September 18, 2026): one credit pool across the web app and Runway Dev, per Runway; new enterprise sign-ups by December 31 get 10% more credits on signing, and existing enterprises that refer another get 5,000 credits plus a day with an Applied AI Architect
• MCP and plugin surfaces (September 17–23, 2026): per Runway’s changelog — a Runway plugin inside Grok Bot (September 17); four Kling models on Runway MCP, Kling O3 4K, Kling 3.0 Standard, Kling O3 Standard and Kling 3.0 Motion Control (September 18, paid plans); Brand Kits readable by MCP agents, read-only, with creation and editing still in the app (September 22); a DaVinci Resolve plugin on paid plans (September 23), the third editing suite after Premiere Pro and After Effects; and Runway MCP listed on Cursor’s Marketplace (September 23)
• Draft mode for Seedance 2.5 (September 24, 2026): per @runwayml, cheaper, faster drafts while exploring, then chosen takes enhanced to full quality — Runway’s surface for ByteDance’s own two-step Draft mode (see Seedance 2.5 under What’s Coming). Runway’s changelog added “Seedance 2.5 Draft Mode in Agent” on October 2: Agent generates fast 480p drafts with post-generation enhancement, suggested automatically and toggled on or off.
• OpenAI Marketplace (September 29, 2026): Runway joined OpenAI’s enterprise Marketplace as a “launch partner” — its words — so eligible enterprises can apply part of an OpenAI spending commitment to Runway Creative (Gen-4.5, Seedance 2.5, GPT Image 2.5 and ElevenLabs V4 listed). OpenAI’s directory lists Runway without that label and says the customer pays the partner directly; no price or revenue share is published (Weekly Roundup — October 5, 2026)
• Runway Ads (September 30, 2026): a product that generates video and image ad creative from a brand kit, publishes approved variants to Meta, Google and TikTok, reads performance back and generates the next round; human approval is on by default. It is piloting with select enterprise partners (early access through Enterprise Sales). Runway’s results — weekly ad volume from 77 to about 900 since July, conversion up roughly 34%, cost per subscriber down 41% — come from its own ad programme, with no outside check
• Changelog, September 29 – October 2, 2026: per Runway’s changelog — ElevenLabs V4 (September 29); Runway inside OpenAI’s Dots (October 1), where a brief given to a dot the night before returns a finished spot in the morning; and, on October 2, Seedance 2.5 Draft Mode in Agent and Ideogram 4.5 in Runway MCP (an image model)
• Grok Imagine Video 1.5 Lite on Runway Dev (October 1, 2026): model ID grok_imagine_1_5_lite; text-to-video and first-frame image-to-video, 1–15 seconds at 480p, 720p or 1080p (Runway notes the 1080p output is upscaled from 720p), billed at 2, 3 and 14 credits per second plus 1 credit for an image-to-video start frame, per the API changelog. Reference inputs go to the full grok_imagine_1_5. Model detail under Grok Imagine.
• Project Continuum (October 1, 2026): @runwayml introduced a research application from Runway Labs built around real-time video interfaces, with four demos — Portals, Visual Thinking, Responsive Video Interfaces and Interactive Worlds — framed as “Interface World Models, like Solaris” (see the Solaris bullet above). No availability, price or named model; the runway.com pages for it returned 404 on October 4. X-only
• Note: Characters is an enterprise API product built on GWM-1, separate from the Gen-4 Turbo generation pipeline

Runway leads in non-photorealistic and stylized video — VFX-oriented aesthetics, abstract content, and artistic directions where other models default to photorealism. Gen-4 Turbo has the most mature professional ecosystem with motion brushes, scene consistency tools, and a robust API. Runway closed a $315M Series C in February 2026 at a $5.3B valuation.

In March 2026, Runway launched Characters — a real-time video agent API built on its GWM-1 world model (Weekly Roundup — March 13, 2026). Characters generates fully conversational AI avatars from a single reference image with no fine-tuning required. The avatars sustain realistic lip-sync, facial expressions, eye contact, and gesture across extended multi-minute conversations, running at 24fps at 720p in real time. BBC and Silverside are early enterprise partners.

At NVIDIA GTC in March 2026, Runway demoed a research preview of a new real-time video generation model (Weekly Roundup — March 20, 2026) running on NVIDIA Vera Rubin hardware — achieving time-to-first-frame under 100ms for HD video. Gen-4.5 became accessible via Adobe Firefly’s multi-model video hub in April 2026 (Weekly Roundup — April 27, 2026) — Runway’s first major distribution beyond its own platform.

In May 2026, Runway launched Runway Agent (Weekly Roundup — May 18, 2026) — a conversational creative partner that runs ideation, generation, sound design, and editing end-to-end from a chat interface. The same month, Runway opened a Tokyo office and committed $40M to Japan, its third-largest market, with its enterprise base tripled in twelve months.

In May 2026, Runway released Aleph 2.0 (Weekly Roundup — May 25, 2026) — an upgraded video editing model that propagates a single-frame edit across the rest of a clip while preserving everything else. Multishot sequences up to 30 seconds at 1080p, edited across cuts in one pass instead of shot-by-shot. Available on all paid Runway plans on the desktop web app.

In July 2026, Runway launched Runway Dev (Weekly Roundup — July 13, 2026) — a developer and enterprise platform putting one API over Runway’s own frontier models (Gen-4.5, Aleph 2.0, Act-Two) plus third-party models including Seedance, GPT Image 2, and ElevenLabs. Recipes package prompting and workflow expertise into single API calls; Workflows chain multiple models into custom pipelines. Runway says it already runs production for Adobe, ElevenLabs, Shutterstock, Figma Weave, Gamma, and Silverside, with SOC 2 Type II compliance and IP indemnification (company-stated). It gives the aggregator strategy a monetization shape — a business product, not just a creator UI.

Later in July, Runway pushed the aggregator play from hosting to routing: Media Router (July 23, 2026) lets a team define what “best” means once — cost, quality, or latency — and automatically selects the video, image, or audio model per request, live in Runway Dev (Weekly Roundup — July 27, 2026).

Pika 2.5 / Pika Agents — Pika Labs

Best for: Budget-conscious creators, rapid iteration, social media content; multi-model agent orchestration via Pika Agents

• Max resolution: 1080p (Pika 2.5); 480p real-time (PikaStream 1.0)
• Max duration: 42 seconds (clip); persistent for live (PikaStream)
• Audio: Supported (Pika Video native; ElevenLabs / MiniMax / OpenAI Whisper via Pika Agents)
• Key feature: Pikaswaps, Pikaffects, fast batch generation (Pika 2.5); PikaStream 1.0 for live agent video; Pika Agents for multi-model orchestration over Kling, Veo, Seedance, MiniMax, and Sora
• Access: Free tier, then Starter $10/mo, Creator $35/mo, Fancy $95/mo and Enterprise, per pika.art/pricing fetched September 20, 2026 — “All models” on every paid tier at 60 credits per dollar (Superseded 2026-09-20: this line read “From $8/mo (lowest entry price among major models)” until the September 17 relaunch repriced the tiers); Pika Agents available at pika.me and across 17+ platform surfaces
• The new Pika (September 17, 2026): relaunched as “an AI creative platform made for and by Creatives,” per @pika_labs; the model menu on pika.art now lists Seedance 2.5, GPT Image 2.5, Grok Imagine 1.5, Gemini 1.1 Omni Flash, Nano Banana 2 and MiniMax H3 alongside Pika 2.5. Third-party write-ups describe automatic per-job model routing and new Swap Anything and Camera Director tools; no Pika primary confirms those, and Pika’s own blog had not posted the relaunch by September 20, so they are not carried here (Weekly Roundup — September 21, 2026)
• Seedance 2.5 Draft Mode (September 23, 2026): on Pika and the Pika API Club, drafts “as low as 10 cents per second,” then finalized at full quality, per @pika_labs. On September 24 Pika offered its API to Grok Bots with a 120-plus-model claim
• API: Available

The most accessible entry point to AI video generation. Pika’s strength is speed and volume — generate 20-30 variations of a concept in minutes, then refine. Features like Pikaswaps (face/object replacement) and Pikaffects (style transfer) add creative flexibility at a price point that undercuts every competitor.

In April 2026, Pika launched PikaStream 1.0 — a real-time AI video engine for live agent meetings at 24fps/480p with ~1.5s speech-to-video latency and persistent identity across calls.

In late April 2026, Pika reintroduced its product line as Pika Agents — a multi-modal AI creative partner that orchestrates other companies’ video models from a conversational interface. The video roster includes Pika’s own model alongside ByteDance’s Seedance 2.0, Kuaishou’s Kling, MiniMax, Google’s Veo 3, and OpenAI’s Sora. On audio: ElevenLabs, MiniMax Music and Voice, OpenAI Whisper. On images: Gemini, ChatGPT Images 2, SeedDream. The agents run inside Slack, Telegram, WhatsApp, Discord, X, Notion, GitHub, Figma, and a dozen other surfaces with persistent memory and personality across sessions.

In July 2026, Pika shipped Director’s Suite (Weekly Roundup — July 13, 2026) — an experimental AI timeline editor for agent-driven video (storyboarding, clip generation, and chat/voice-guided editing in one interface), alongside a VFX Skill and a Seedance-4K-powered 4K-VFX Skill that edits an existing clip from a single prompt while preserving faces, gestures, audio, and camera moves.

MiniMax H3 — MiniMax

Best for: Omni-modal generation and editing at the value tier — 2K with native stereo audio, priced and benchmarked from day one; the first arena leader in generative video editing

• Max resolution: 2K (2560×1440)
• Max duration: 5–15 seconds
• Frame rate: 24fps
• Audio: Native stereo audio
• Arena (October 11, 2026): #4 text-to-video v2.0 with audio (Elo 1,137 at 768p) and #3 silent (1,117); #2 image-to-video (Elo 1,181); #4 video-editing with audio (Elo 1,119) — H3 held #1 on video-editing from its August 3 debut until Wan 3.0 displaced it on August 19, and Seedance 2.5 and Omni Flash 1.1 now sit above it. On image-to-video MiniMax H3 Max, a third-party post-train of H3’s own open weights, sits above it; on text-to-video v2.0 H3 is a place ahead of H3 Max with audio and a place behind it silent, inside the error bars on both, behind a second post-train, Utopai X, with audio
• Key feature: Omni-modal input (text/image/video/audio references, up to 9/3/3 respectively); in-place character, scene, dialogue, and voice editing; the video-editing arena win
• Access: Consumer via the Hailuo AI app; API live day one
• API: $0.13/sec at 2K, model ID MiniMax-H3
• Weights: Shipped August 3, 2026 (H3-Base, 33.1B-parameter dense omni-modal transformer, day-one ComfyUI/vLLM-Omni/SGLang support) — but the Community License excludes the US, EU, UK, and South Korea from its “Applicable Territory” entirely, and local self-hosted generation caps at 768p, not the API’s 2K — the 2K upscaling module isn’t open-sourced and still routes through MiniMax’s hosted infrastructure
• Caution (license): The Community License grants nothing in the US, EU, UK or South Korea: use, display or distribution of the weights, their derivatives or their Outputs outside the Applicable Territory is not authorized (§ V.4). MiniMax invites anyone in an excluded territory to ask for a separate license (§ II), and requires separate written authorization from commercial products above $20 million in yearly revenue (§ IV.1).
• Caution: Not on the Commission’s list of signatories to the EU’s Article 50 transparency Code of Practice (list read October 11, 2026), as the Article 50 obligations took effect August 2, 2026. Of the labs this page tracks, Google, Meta, Black Forest Labs and OpenAI are on it.

MiniMax launched H3 — successor to its Hailuo 2.x line — on July 31, 2026, and debuted directly into the field’s top tier: top-three finishes on both major generation boards, and an outright #1 on Artificial Analysis’s newer video-editing leaderboard, a blind-preference category the T2V and I2V boards don’t test. It’s the strongest arena debut by any non-Google lab this page tracks.

The launch posture stood out on its own terms too. Where ByteDance’s Seedance 2.5 went “official” the same week without a published price (see Seedance), H3’s API was live and priced from hour one. MiniMax also kept its open-weights promise fast — weights landed August 3, roughly 60 hours after the “within days” pledge, the first credible challenger to HappyHorse’s long-unfulfilled open-weights claim in this same tier. But “open weights” comes with real fine print: MiniMax’s own license withholds the US, EU, UK and South Korea from the grant entirely (while inviting separate requests from there), and the self-hosted build doesn’t reach the API’s 2K ceiling. Read the claim precisely — H3 is genuinely open-weights somewhere, for some resolution, which is still more than HappyHorse has delivered anywhere.

Then the open weights did the thing open weights are supposed to do. In late August, fal Research shipped H3 Max, a post-train of H3 that took image-to-video outright and, on the rebuilt text-to-video v2.0 board, sits a place behind the official endpoint with audio and a place ahead of it silent, inside the error bars on both toggles. MiniMax’s own account endorsed the result, calling it “exactly why we build with open weights.”

The pattern kept going in September. fal shipped H3 Max Turbo and H3 Max Director on the same weights (September 2 and 3), and on September 2 a research team unaffiliated with either company — Xingyi Yang, PolyU/NUS/Oxford — published H3-World, converting the open H3 into a world model on 8,000 training samples with 0.199% of its parameters trainable. No product, no independent benchmark, and the fifth outside build on one open release in the five weeks since it shipped (Weekly Roundup — September 7, 2026). Two days later Viggle shipped Viggle-Animate (September 4), which it calls its first open-weight model: a post-train of H3 that replaces a character in a video from a single repainted frame — no pose skeleton, no mask — at 124 frames in 26 seconds on one B200, per the model card. The same card states the weights are a Model Derivative of MiniMax H3 under the MiniMax H3 Community License, which carries the territory clause above; Viggle’s announcement does not mention it. RCTV’s open-weights claims tracker logged it as the first installment’s lead case, beside HappyHorse’s weights that never shipped and the territory clauses on H3 and HY-World 2.0. Sixth outside build, and the first to inherit the fine print without saying so. Video Rebirth open-sourced HyperFlow on September 24: an 8-step LoRA for H3, obtained by data-free flow self-distillation, that replaces the default 49 model forwards with 8. Its model card states the Community License’s territorial scope in plain terms; the announcement post does not. Seventh outside build, and the second whose announcement leaves the fine print to the card. On September 30 three more arrived in a day. HeyGen’s HeyGen Video, its first general-purpose model (text, image and reference to video, with sound), says on its own catalog page that it is built on MiniMax H3 and post-trained by HeyGen, at a promotional $0.01 per second through October (the page’s own list-price figures disagree), through HeyGen’s API, OpenRouter, Runware and ComfyUI; its quality figures are HeyGen’s own blind tests, and it is on neither Artificial Analysis board. Creatify’s Boreal-H3 is post-trained “specifically for advertising,” with vendor-reported gains (85.3% reference fidelity, brief success from 28% to 50%) and no stated baselines; it is not the earlier “Boreal” on Artificial Analysis’s board. Utopai X, Utopai Studios’ model inside its PAI platform, debuted second on the rebuilt text-to-video board (1,149 as of October 11) with no public API; Artificial Analysis describes it as post-trained on MiniMax H3, while Utopai’s own launch post names no base model. Eighth through tenth outside builds. MiniMax’s account quote-posted HeyGen’s launch, saying it was proud to provide the foundation, and welcomed Creatify’s. The license they inherit excludes the United States; HeyGen is a US-established company with a Los Angeles address and Utopai is headquartered in Mountain View, and whether either holds a separate license from MiniMax is not stated anywhere we can read (Weekly Roundup — October 5, 2026).

Wan 3.0 — Alibaba

Best for: Long single-pass takes and generative video editing — the current #1 on the Artificial Analysis video-editing board with audio and on its rebuilt text-to-video v2.0 boards

• Max resolution: 1080p (480p / 720p / 1080p tiers; no 4K)
• Max duration: 2–30 seconds in a single pass at 30fps, no chaining and no stitching; a -1 duration hands the model smart duration selection (default 5s)
• Key feature: Document-to-video — accepts docx/doc, xlsx/xls, pptx/ppt, pdf, txt, key, pages, numbers and md, ≤100MB and ≤50 pages, as generation source material. Reference-to-video takes up to 10 images, 5 video clips (≤15s total) and 5 audio clips (≤15s total). Two model IDs: wan3.0-video and the speed-tier wan3.0-video-prime
• Arena (October 11, 2026): #1 on video-editing with audio (Elo 1,187; second to Omni Flash 1.1 without audio, 1,146 against 1,157) and #1 on text-to-video v2.0 — 1,156 with audio (rank range 1–2, level with Utopai X) and 1,129 silent; #7 on image-to-video (Elo 1,164), the one place it trails clearly. The September text-to-video lead swaps with Omni Flash were on the retired v1 scale; on v2.0 Omni Flash 1.1 is eighth. It displaced Omni Flash and MiniMax H3 on text-to-video and video-editing on August 19 and has held video-editing with audio since. Also debuted #1 on arena.ai’s separate Video Edit Arena
• Access: Invite-gated beta opened August 6, 2026 via Alibaba Cloud Model Studio (Beijing and Singapore endpoints) and Qwen Cloud. Label conflict, logged: trade coverage on August 24–25 described an official release, while Alibaba’s own Model Studio API reference still read “Currently in preview.” This page tracks it as preview until the vendor docs say otherwise — vendor documentation governs for procurement. Re-checked 2026-09-01 against a page Alibaba itself stamped “Updated at: 2026-09-01”: it still reads “Currently in preview.” Six endpoint regions, not five — Singapore, Beijing, Hong Kong, Tokyo, Frankfurt and US Virginia — with cross-region calls failing outright. Also live on Runway: the consumer app since August 24, the Dev API since August 26, billed at 5/10/20 credits per second for 480p/720p/1080p. This is the gap that cost R#19 its discovery score: no monitoring lane covers vendor API docs
• API: List: $0.05/sec (480p) · $0.10/sec (720p) · $0.20/sec (1080p) — so a 30-second 1080p take lists at $6.00. Alibaba’s pricing page, stamped October 3, carries a live limited-time 30% discount on this tier, which puts the same take nearer $4.20 at the currently-billed rate. The separate speed tier wan3.0-video-prime is not discounted: $0.068/$0.14/$0.28 per second. Audio is on by default and, in Alibaba’s own words, “enabling or disabling audio does not affect pricing”
• Weights: None. This is a closed commercial release — and so, it turns out, is Wan 2.7; Alibaba’s open weights stop at Wan2.2 on both github.com/Wan-Video and Hugging Face
• Caution: Invite-gated. A model can lead a blind-preference board and still be something most readers cannot buy

Alibaba opened Wan 3.0’s beta on August 6, 2026 and it charted for the first time on August 19 — absent from every Artificial Analysis board on the August 10, 12, 14 and 17 fetches, then straight in at #1 on two of them. It has held video-editing with audio since; text-to-video was rebuilt on September 30 and it leads the new board.

The interesting part is what it is, not where it ranks. Alibaba already runs the most-downloaded open-weights video line in the field (Wan 2.2, the last open release in the Wan 2.x line); Wan 3.0 is the same company shipping a closed, invite-gated, commercially-licensed model above it. The 30-second single-pass ceiling is the technical claim worth watching — every other model on this page reaches long-form by chaining shorter takes, which is where continuity breaks.

Distribution moved the way it now always moves: Runway on August 24, Pika and ComfyUI within two days, then ArtArch, Media.io, OiiOii AI, Picsart, PowerDirector, A2E, PixelDojo and DeepInfra inside a week — eleven surfaces in eight days, the same absorption pattern already documented for Seedance 2.5, FLUX 3 and MiniMax H3.

MiniMax H3 Max — post-trained by fal

Best for: Speed. A 5-second 768p clip in under 3 seconds — the first model this page tracks that generates faster than the clip plays

• Max resolution: 1080p — fal’s model page now prices a 1080p tier ($0.16/sec list) alongside 480p and 768p. (Superseded 2026-09-13: through September 6 this entry read 768P only, which matched fal’s API schema at the time — 480P and 768P, no 1080p or 2K tier. The tier is new, not missed. The entry briefly carried “1344×768” — an inference no source states; removed 2026-09-01.)
• Max duration: 10 seconds on the free tier; 5-second text-to-video and image-to-video are the standard calls
• Arena (October 11, 2026): #1 image-to-video (Elo 1,195), holding the place it took from Dreamina Seedance 2.0 720p in late August; #5 text-to-video v2.0 with audio (1,130), a place behind the official MiniMax H3 endpoint (1,137, inside the error bars), and #2 silent (1,118, a place ahead of H3’s 1,117)
• Speed claim: fal claims 35× the throughput of the official H3 endpoint; Design Arena independently measured 50×; against a quality-matched peer group the figure is 15×. These are three different comparisons and should never be blended
• Access: Five 5-second generations a day with no account; signing in raises it to 15 a day at up to 10 seconds (raised from 5 on August 31, text-to-video and image-to-video both). Also on Runway Dev since September 3, 2026 (model ID h3_max, 5–15 seconds at 480p or 768p, 5 / 8 credits per second) per Runway’s API changelog — a second vendor distributing a third party’s post-train of MiniMax’s weights
• API: $0.08/sec at 768P ($4.80/min), $0.05/sec at 480P, $0.16/sec at 1080p — list rates, from fal’s own model page. A 40%-off promotional rate now runs to October 15, 2026: $0.048/sec at 768P, $0.03/sec at 480P, $0.096/sec at 1080p, per fal’s model page fetched October 4. (Superseded 2026-09-20: through September 14 the promotion was 75% off — $0.02 / $0.0125 / $0.04 — with an end date fal moved from September 7 to September 14; it has now halved the discount and moved the date a third time. The earlier figures were right when written.) (Superseded 2026-10-04: the 50% rate that ran to September 30 became 40% off to October 15 — the fourth change to this promotion.) Also on Runway’s Fall 2026 collection since September 16 (Weekly Roundup — September 21, 2026), alongside its Runway Dev listing. (Corrected 2026-09-01: this page and R#19 described the promotion as half price for 14 days. It is 75% off.)
• H3 Max Turbo (September 2, 2026): a preview variant fal describes as twice H3 Max’s speed at half the cost, aimed at “the 97th percentile” of the base model’s quality — fal’s own target, unscored by any board, and absent from all three Artificial Analysis boards as of September 13. List $0.04/sec at 768P, $0.025/sec at 480P, $0.08/sec at 1080p from fal’s model page; the launch promotion is now 40% off to October 15 ($0.024 / $0.015 / $0.048, fetched October 4), down from 50% to September 30 and 75% to September 14, moving in step with H3 Max’s own promotion
• H3 Max Director (September 3, 2026): the continuous, action-controlled variant behind fal.live, now a paid API on its own model page — $0.08/sec (fal’s Director page, fetched October 4, still carries a sentence saying a promotion expired September 14, though it no longer quotes a promotional rate), 1080p at twice that, billed at a $4.80 minimum per session (the page read $1.20 on September 27); Director 1.1 (September 8) adds keyframe insertion mid-generation, audio introduced at any point, native 1080p at twice the per-second rate, and sessions up to 15 minutes by default, per fal’s post and the model page
• Multi Angle (September 9) and Reference-to-Video (generally available by September 10): Multi Angle renders a scene from a chosen horizontal and vertical camera angle, set in degrees, from one image in under three seconds; Reference-to-Video moved to general availability on fal. Six H3 Max releases from the August 25 fine-tune to September 9, listed in Weekly Roundup — September 14, 2026
• 3D to Video (September 23) and Extend Video (September 24): 3D to Video turns a Blender previs of up to 15 seconds into photoreal video while keeping its camera movement and shot timing; Extend Video lengthens an uploaded clip of up to 15 seconds to 30. Both on fal, both unscored by any board
• Recast (October 1) and Reference-to-Video frame control (October 2): fal Recast swaps any character in an existing video for a person in a reference photo while keeping the source’s motion, camera moves, cuts and audio, powered by H3 Max; fal’s model page prices it at $0.30 per second of output at 768p and $0.45 at 1080p, reference images free. The next day fal added first-, middle- and last-frame control to H3 Max Reference-to-Video. Both unscored by any board. Recast is a product built on a post-train of open weights by the post-train’s own host, at 3.75 times the 768p list rate of the underlying text-to-video endpoint ($0.30 against $0.08 per second).
• Base model: A post-train of MiniMax’s open-weight H3, developed by fal and run on its own inference stack — not an independent architecture. As a Model Derivative it carries H3’s Community License, which bars using or displaying the model or its Outputs in the EU, the UK, South Korea and the United States; fal, a San Francisco company, has not said publicly whether it holds a separate MiniMax license. On fal’s launch blog, MiniMax’s H3 team calls fal “a natural partner for H3 Max.” (Corrected 2026-10-04: this line said H3 Max carried none of the license’s territory exclusions “because fal serves it hosted.” Nothing supported that, and the license text says otherwise: hosting is not an exemption.)
• Caution: fal’s own board label reads “#1 image-to-video with audio.” That rank is real; see the leaderboard note under Quick Reference for what Artificial Analysis’s audio labelling actually means
• Note on the price we would not print: R#19 deliberately published no per-second rate, because fal’s pricing widget did not render and a summarizing fetch produced figures we refused to launder. The page rendered cleanly on 2026-09-01 and the real list rate is $0.08/sec — the same number the summarizer had guessed. Refusing to print an unverified figure that later proves correct is the process working, not a miss

This is the first time a third-party fine-tune tops a leaderboard this page tracks, and it is worth being precise about what that demonstrates: not that fal built a better model than MiniMax, but that MiniMax’s decision to publish open weights let someone else find headroom in them. MiniMax’s own account endorsed the result rather than disputing it.

The speed number is the story (Weekly Roundup — August 31, 2026). Generation faster than playback is a category threshold, not a benchmark position: it is the precondition for live and interactive video rather than rendered-then-delivered video. fal tested that directly with fal.live on August 31 — interactive AI livestreams driven by viewer prompts, running on “H3 Max Director,” an autoregressive variant holding up to two minutes of native context. fal paused it roughly three hours after launch, citing quality and experience work, without disclosing what broke. It came back in pieces — two of seven channels on September 3, six of nine on September 6, with two channels added, three still paused, and no account from fal of what failed — and Director shipped as an API product on September 3 regardless (Weekly Roundup — September 7, 2026). Treat the threshold as real and the sustained-load behaviour as unproven.

Sora 2 — OpenAI

Status: Discontinued March 24, 2026; consumer app shutdown executed April 26, 2026; API removed September 24, 2026. RCTV analysis →

OpenAI announced Sora’s discontinuation on March 24, 2026 — the app, the API, and the Disney licensing deal announced with it in December 2025. The stated reason was compute reallocation toward “world simulation for robotics.” The numbers tell the fuller story: estimated $15M/day peak inference cost against $2.1M in total lifetime in-app revenue, and a 66% download decline from its November 2025 peak to February 2026. Sora is removed from active tracking. See Weekly Roundup — March 27, 2026 for the full breakdown.

Shutdown timeline: The Sora consumer app and web interface went dark on April 26, 2026 (Weekly Roundup — April 27, 2026) — the export window closed at that time. The Sora Videos API — sora-2, sora-2-pro and every dated snapshot — was removed on September 24, 2026, per OpenAI’s API deprecations page, which records the developer notice as March 24 and lists no recommended replacement for any Sora row; the sunset FAQ offers export, refunds and data deletion, nothing else. That closed the second and last stage of the shutdown: OpenAI exits the video-generation infrastructure layer as well as the consumer product (Weekly Roundup — September 21, 2026).

The Agentic Layer: Orchestration on Top of the Models

The model is no longer the product. The six agents in this section don’t generate video — they decide which model generates video, and on what schedule, across what surfaces. Think of the commercial models above as engines: Veo 3.1, Kling 3.0, Seedance 2.0. The agents are steering wheels. The operator-relevant question has shifted up a layer: not “which model is best?” but “which agent puts the right model on the right shot, in the right workflow, with the right context carried forward?” That question didn’t exist twelve months ago. Six companies have already shipped an answer.

Luma Agents

Luma AI shipped the first production-grade conversational agent in this category in March 2026. By March 10, deployments were live at Publicis Groupe and Serviceplan Group — no beta waiting period. The agent works from a Luma Uni-1 reasoning layer that plans and coordinates across video, image, audio, and text before generating anything. It calls Luma’s own Ray 3.2 / Ray 3.14 for video, Google’s Veo 3 for photorealistic shots, Nano Banana Pro, ByteDance’s Seedream, and ElevenLabs for voice. Seedance 2.0 joined the roster in May 2026. The Mazda commercial deliverable in April 2026 — Johannesburg agency Boundless produced Mazda’s first AI-generated commercial using Luma Agents in under two weeks — is the most credible production-deployment signal any AI video agent has produced.

The critical technical feature: persistent context across the full asset suite. Luma Agents remembers what was generated in earlier steps and can revise upstream elements when downstream evaluation surfaces a problem. That is what makes it an orchestrator rather than a fancy prompt box.

Pika Agents

Pika’s April 28, 2026 launch was the moment the agentic-orchestration pattern became industry news rather than one lab’s experiment. Pika Agents orchestrates a broader model roster than any competitor — Kling, Veo 3, Seedance 2.0, MiniMax, Sora’s API, and Pika’s own model on the video side; ElevenLabs, MiniMax, and Whisper on audio — all from a conversational interface that runs inside 17 surfaces where creators already work. RCTV covered this as the R#2 lede because the framing was explicit: “the prompt is no longer the product.” That line has since become the editorial spine for this entire category.

PikaStream 1.0 (April 2, 2026): the real-time video engine that runs inside Pika Agents as its live-avatar capability — 24fps at 480p, ~1.5-second speech-to-video latency, persistent identity across calls. It is a streaming runtime, not a standalone orchestrator; the agent layer above it is Pika Agents.

Runway Agent

Runway shipped its agent in May 2026 — single-vendor, full Runway stack. The positioning is end-to-end: Runway Agent handles concept ideation, generation via Gen-4 Turbo or Aleph 2.0, sound design, and editing inside the same conversation thread. Runway’s “single-vendor” constraint is a deliberate product posture, not a technical limitation — they own the generation and editing stack, so the agent never needs to leave it. Whether that narrows or focuses the use case depends on whether the operator’s workflow already lives in Runway. Available on all paid plans, starting at $12/month.

Runway added an enterprise anchor and a new capability layer in July 2026. On July 1, Runway announced a partnership with Bertelsmann, integrating its models “across Bertelsmann’s global portfolio of businesses” — RTL Group, BMG, and Bertelsmann Marketing Services. RTL’s Fremantle already runs Runway through its in-house AI studio, Imaginae; BMG uses it for artist marketing visuals — a genuine enterprise-adoption story, not a single pilot. The next day, Runway shipped Agent Skills: a slash-command layer on top of Runway Agent for building ad campaigns, commercials, and localized ads on demand. Runway is now the third vendor, after Pika and Luma, to converge on “Skills” as the name for a portable, reusable agent workflow.

In July 2026, a new third-party benchmark for AI video agents — Physion-Arc 1.0 from Physion Labs — ranked Runway Agent 2.0 #1 of six agents tested (against Luma, MiniMax, Kling, Utopai Studios, and TapNow), leading every dimension it scored: narrative coherence, cinematic language, production quality. The caveats are load-bearing: the benchmark is brand-new, its methodology page is unreachable, its two primaries disagree on the test’s own scale (Physion Labs cites 100 screenplays / 600 videos; Runway’s amplification cites “30 cinematic prompts and 16 metrics”), and Runway is the loudest amplifier of a result that crowns Runway — no independent replication yet (Weekly Roundup — July 27, 2026).

Higgsfield Supercomputer

Higgsfield’s Supercomputer, launched in mid-May 2026, is the most enterprise-positioned agent in the set. “Supercomputer” is the framing: orchestrate Claude, GPT-5.5, and Gemini models alongside Higgsfield’s video stack to plan and execute full content campaigns end to end. On May 19, 2026 — the same day as Google I/O — Higgsfield updated the orchestration layer to Gemini, describing the swap as “8× cheaper, 3× faster.” On May 29, Higgsfield upgraded its Claude backbone to Opus 4.8, and on May 30 shipped Higgsfield Reframe — an MCP-native aspect-ratio reframing tool available inside Claude. The multi-model roster is the broadest LLM coverage in this category. Distribution extends via browser, Telegram, and 30+ third-party integrations. In a 12-day sprint through early June 2026 (Weekly Roundup — June 15, 2026), Higgsfield added five external surfaces — Claude MCP (May 28), Adobe Premiere/After Effects plugins (May 29), Figma (June 4), a Minecraft mod (June 5), and a DaVinci Resolve plugin (June 8) — and folded Grok Imagine 1.5 into its own platform, the clearest expression yet of the routing bet: a competitor’s model generating video inside Higgsfield’s surface. The pipeline expanded again at the end of June and start of July: audio generation landed in the MCP on June 27 — voiceovers, voice cloning, and dubbing across 50-plus languages via Seed Audio 1.0 and ElevenLabs v3 — and on July 2 Higgsfield shipped Explainer, generating faceless documentaries up to 10 minutes from an auto-researched script, built on Claude Fable 5 plus Gemini Omni Flash. The same day, Higgsfield made its MCP free: 100 credits and a 3-day full-access trial, removing the price floor entirely two days after Google’s own Omni Flash preview shipped. On July 3, Higgsfield pushed into the professional edit suite itself — plugins bringing Gemini Omni Flash and Seed Audio 1.0 into Adobe Premiere Pro and DaVinci Resolve Studio timelines, for background cleanup, multi-shot generation, and 18-language dubbing without leaving the NLE. Higgsfield’s Supercomputer page carries the full capability description.

The round closed. August 17, 2026: $400M Series B at a $5.4B valuation — DST Global led, with Goldman Sachs Alternatives, Valor Capital and Tribe Capital participating. That is roughly 4× the ~$1.3B mark of eight months earlier. Disclosed alongside it: $700M annualized revenue, 30M users across 200 countries, and 390 Fortune 500 enterprise clients. (This page tracked the round in July as “reportedly in talks to raise $300–500M at a $5B valuation” on a ~$500M run rate; the closed terms came in above the reported range on both figures.) CEO Alex Mashrabov frames the raise as compute rather than growth capital — “Video is one of the most compute-intensive domains in AI,” putting one minute of video at the resource cost of 60,000 words of text. The revenue trajectory is still the more interesting number than the valuation: ~$200M at the end of 2025 to $700M annualized eight months later, in a category where nobody else discloses revenue at all (Weekly Roundup — July 27, 2026).

Higgsfield API (announced September 16, live September 17, 2026). The launch post pitches “50+ frontier models in one API, at lower prices than a subscription” — pay per use, no commitment, up to 50% off a user’s three chosen models if the discount is locked within seven days. The catalog on Higgsfield’s own API console listed 22 models on September 20, 14 of them video — Seedance 2.5 (30% off), Kling 3.0 (50% off), MiniMax H3 (45% off), Wan 3.0 and Wan 3.0 Prime, Seedance 2.0, Grok Imagine Video 1.5, LTX-2.5 Fast and Pro, Wan 2.7, HappyHorse 1.0 and 1.1, plus Higgsfield’s own Genjutsu and Cinema Studio 4.0 — and eight image models. docs.higgsfield.ai still documents only the Soul family with no September changelog, so what Higgsfield counts toward “50+” is not stated anywhere it publishes. Carried as the vendor’s claim beside the count visible on its own surface (Weekly Roundup — September 21, 2026).

$1B annualized revenue, self-reported (September 24, 2026). Higgsfield says its annualized revenue crossed $1 billion 18 months after launch; Bloomberg’s headline has it eyeing $1 billion rather than past it. Unaudited either way. The documented trajectory is roughly $200M at the end of 2025 to $700M at the August 17 close (Weekly Roundup — September 28, 2026).

Seedance 2.5 Preview Mode and API cashback (September 23–27, 2026). Preview Mode (September 23) renders a 480p draft, then the chosen take at 1080p. On September 24 Higgsfield announced 100% instant cashback on API spend across every model on the platform, up to $100,000 per business, from a $20 million pool, first come, first served. Posts from September 26 onward pair it with native-1080p Seedance 2.5 on the API, raise the ceiling to $200,000 and describe the cashback as API credits, with $18 million left in the pool. The September 26 post added that unused cashback expires September 30. Higgsfield’s API console, read October 4, now describes only a standing monthly programme — 5% back at $5,000 of spend, rising to 20% at $50,000 — and says nothing of the 100% launch rebate, so what happened to that pool is not stated anywhere we can read. It was a rebate in credits, not a price cut.

Adobe Firefly AI Assistant

Adobe is the incumbent here, and the only one that arrived via acquisition of creative-workflow context rather than from a blank slate. Firefly AI Assistant — previewed as Project Moonlight at MAX, public beta April 27, 2026 — orchestrates the Creative Cloud stack conversationally: Premiere Pro with full project metadata access, Photoshop, Lightroom, Illustrator, Express. Creative Skills are the execution layer — predefined multi-step workflows that fire from a single natural-language instruction. The operative claim is multi-app handoff without context loss; the operator verdict is still forming in the public beta.

Generative Media Tool in Premiere Pro and After Effects (September 8, 2026). Adobe put generation on the timeline itself: a native tool that generates video and sound effects in place, filling gaps from reference frames sampled from the project, with a model picker offering Firefly and, in Adobe’s words, “partner models including Google Veo, Kling, Runway and Luma.” Adobe’s post says nothing about how partner-model generation is billed. It is the multi-model hub reaching the edit, and it arrived the same day as Runway’s own plugin into the same two apps with a partly different roster (Weekly Roundup — September 14, 2026).

The AI-native vs. incumbent frame is worth naming: Luma, Pika, Runway, and Higgsfield are all building agents on top of AI video first. Adobe is extending an agent over a suite it already controls. The operator question these two approaches answer is different — and which model wins the relationship depends on whether the operator’s workflow is already Premiere-centric or starting from scratch.

HeyGen HyperFrames

HeyGen shipped HyperFrames “Next Generation Skills” on June 26, 2026 — nine production workflows (launch video, music video, captions and overlays on video, and six more) routed by context rather than by menu. HeyGen’s own framing: “it knows which one you mean from context and routes there on its own.” There’s no workflow picker to click through; the agent infers intent from the prompt and dispatches to the right pipeline. At 1.53 million impressions, it’s the highest-engagement AI-video product launch RCTV tracked in the June coverage window — a genuine adoption signal, even accounting for the platform’s own promotional push. HyperFrames is a single-vendor agent, orchestrating HeyGen’s own production models rather than a third-party roster — closer to Runway Agent’s posture than to Pika Agents’ multi-model bet. Surfaces at hyperframes.heygen.com. In July 2026, HyperFrames product engineer James Russo detailed the architecture in an X-Article, “HTML Is All Agents Need” — HyperFrames renders agent-authored video from plain HTML/CSS/JS rather than a proprietary timeline format, on the reasoning that the model already knows HTML fluently. Cited traction: 1.3M videos for 267K creators in 90 days, 32K GitHub stars, open source, running inside Claude Code, Codex, and Cursor (Weekly Roundup — July 13, 2026).

MCP server inside Claude Design (August 7, 2026). HyperFrames shipped an MCP server that puts the whole assembly step behind a share action: animate a design, hit share, and a HyperFrames agent scores the result, adds sound effects, tightens the motion, and returns a finished MP4 — “Just send it to HyperFrames. We’ll take it from there,” in the vendor’s own words. No human touches the step between design and shippable video. This is the agentic-assembly pattern reaching a design surface rather than a video platform (Weekly Roundup — August 17, 2026).

LiveAvatar concurrency limits removed (August 25, 2026). HeyGen dropped the cap on concurrent LiveAvatar sessions entirely — “run 1 avatar or 10,000 at once, same API, full-body 1080p, down to $0.01/min at scale.” A capacity and pricing change rather than a new model, but the direction is worth noting: the constraint on real-time avatar deployment moves from what the vendor will allow to what the customer will pay.

LiveAvatar and HyperFrames on GPT-Live-1, open-sourced (September 10, 2026). HeyGen published a demo stack pairing its real-time avatar renderer and HyperFrames’ live on-screen visual generation with OpenAI’s newly released GPT-Live-1 full-duplex voice model, code public on GitHub (heygen-com/liveavatar-gpt-live-demos). The voice model is OpenAI’s news; the part that belongs on this page is motion graphics appearing mid-sentence, generated by the agent doing the talking.

Code2Video benchmark (report dated September 18; launched on X September 21, 2026). HyperFrames’ first benchmark, built with Google DeepMind and Kaggle: 168 human-authored motion-graphics briefs drawn from product-launch videos, language models write HyperFrames code for each, and a purpose-trained judge model scores the rendered result on five axes — Engagement, Prompt-Intent, Composition, Temporal and Craft. The judge agrees with human raters 82% of the time against 75% for general vision-language models, rising to 98.8% on its most confident calls, per HeyGen’s report. The top four models sit within 12 Elo of each other with overlapping intervals, open-weight models land within 50 Elo of first, and the spread is narrowest on Prompt-Intent and widest on Composition and Craft. The artifact is code-rendered motion graphics, not diffusion video: it scores the assembly layer this entry covers, not a generation model (Weekly Roundup — September 28, 2026).

HeyGen also shipped HeyGen Video, a general video model post-trained on MiniMax H3 (September 30); see H3.

What to watch

Two August developments sit directly on these axes. ComfyUI shipped Comfy MCP local and open-source on August 18, 2026 — “the #1 ask after Comfy Cloud MCP shipped in June.” An agent now sees a user’s actual local install (every custom node, every model on disk), can fetch files, start the instance, and get a workflow running end to end. The self-hosted layer now has the same agent-driven entry point the hosted platforms have been building all year, which is the first time this category extends below the API line. Separately, Tencent Hunyuan previewed HyCreator (August 20–21, 2026), an agent harness claiming an “End2end Auto Mode” that generates 10-minute-scale films with “zero human intervention,” alongside a real-time interactive-editing mode; early access is application-gated. The claim is far ahead of anything demonstrated in this section and is carried here as a claim, not a capability — no independent hands-on exists yet.

ComfyUI added a routing layer and a deployment layer in the same week. Comfy Router (September 23) puts frontier image, video, 3D and audio models — Seedance 2.5, MiniMax H3, Kling and Black Forest Labs models at launch — behind one API; where a model runs on more than one of fal, Runware, WaveSpeed and Higgsfield the provider is a parameter, a named provider is never silently swapped, and billing is in Comfy credits with no subscription. Comfy API (September 30) is the other half: a workflow packaged with its nodes, models and dependencies deploys as an autoscaling endpoint, live on paid Comfy plans. The self-hosted layer is now a storefront too.

fal Agent (fal.ai/agent) is a conversational creative agent with project memory that picks the model per job; pay-as-you-go or $50 / $200 / $1,000 monthly credit plans. It passes this section’s test (conversation, persistent context, several models), so it is a seventh agent in all but the table, which waits on a structural decision.

Claude Motion (Anthropic, October 8, 2026) is a beta on Team and Enterprise plans that turns a prompt into a short animated explainer and exports an MP4. Anthropic says it ‘doesn’t use a video generation model’: Claude writes code that animates the user’s own text, charts, shapes and images, so there is no generated footage and no AI-generated people, the same code-rendered pattern as HyperFrames above. Its tie to this page is the handoff: Anthropic lists Adobe, Descript, HeyGen, Higgsfield, invideo, Luma AI and Runway as places to open the animation next, and Runway calls itself a launch partner and pitches generated footage added around the animation. A distribution surface for tracked tools, not a generation model (Weekly Roundup — October 12, 2026).

Three axes will decide how this layer develops. First: single-vendor vs. multi-vendor convergence. Runway’s Runway-only posture contrasts with Pika’s eight-model roster; whether Runway opens to third-party models is the specific question. Second: AI-native startup agents vs. legacy-incumbent agents. Adobe’s Creative Skills framework is embedded in the world’s most-used professional NLE; Pika Agents runs in Slack. Neither is clearly winning the operator relationship yet. Third: agent-to-agent interoperability. None of the six currently calls another vendor’s agent — they call models. The day Pika Agents calls Runway Agent, the competitive dynamics of this category change entirely.

Open-Source & Local Generation

The open-source AI video ecosystem has matured significantly, making local generation on consumer hardware a viable option for privacy-conscious creators and developers.

LTX-2.5 — LTX

Best for: Local/desktop generation, consumer GPU workflows, high-frame-rate output

• Vendor: LTX — an independent company since July 8, 2026, spun out of Lightricks (which keeps its consumer apps, e.g. Facetune); CEO Zeev Farbman
• Current release: LTX-2.5 (August 11, 2026). The spec bullets below are the verified LTX-2.3 card; a full 2.5 spec card is pending primary verification of resolution, duration, and frame-rate figures — the vendor’s launch post does not state them.
• Max resolution: 4K native (true 4K, not upscaled)
• Max duration: 20 seconds
• Frame rate: Up to 50fps (24/48fps options also available)
• Audio: Native synchronized audio (improved HiFi-GAN vocoder)
• Portrait mode: Yes (9:16, up to 1080×1920)
• Hardware: Runs on GPUs with 12GB+ VRAM; optimized for RTX 50 Series (2.5× faster via NVFP4)
• Integration: ComfyUI native; standalone desktop video editor (shipped March 2026)
• License: the LTX-2.x Community License Agreement (dated August 11, 2026) — open weights, but not an OSI-open licence. (Corrected 2026-09-01: this page said “Apache 2.0” in three places while its own licence bullet described a $10M revenue threshold Apache 2.0 cannot contain. Verified against the licence text itself.) Free commercial use below $10,000,000 annual revenue, aggregated across affiliates under common control; at or above, a separate paid Commercial Use Agreement is required. Four restrictions worth knowing before you build on it: derivatives are defined broadly — fine-tunes, adapted weights, distillation via intermediate representations, and training on LTX-2.x-generated synthetic data all count; commercial use may not train a competing model on the weights or their outputs; military, weapons and nuclear applications are barred; and disabling or circumventing watermarking or provenance functionality is barred in two separate clauses. LTX states it intends the model to qualify as free and open-source under EU AI Act Article 53(2), with an explicit carve-out that the derogation does not extend to Article 53(1)(c) and (d)

A comprehensive rebuild released in March 2026 (Weekly Roundup — March 20, 2026): a new VAE for sharper detail, a 4× larger text connector for better prompt understanding, and an improved HiFi-GAN vocoder for cleaner native audio. The model ships alongside a dedicated desktop video editor, making the entire local pipeline accessible without a ComfyUI node graph.

Key capabilities: native portrait mode (9:16 up to 1080×1920), last-frame interpolation for seamless clip chaining, and 24/48fps output options. At GDC 2026, NVIDIA announced 2.5× performance gains on RTX 50 Series via NVFP4 quantization, 60% lower VRAM usage, and RTX Video Super Resolution for ComfyUI delivering 4K upscaling 30× faster than competing local alternatives. The ComfyUI App View strips the node-graph interface into a simplified prompt-in/video-out UI for non-technical users.

LTX became its own company, then shipped LTX-2.5 (August 2026). On July 8, 2026 LTX spun out of Lightricks as an independent open-world-models company, taking the video line and CEO Zeev Farbman with it; Lightricks retained its consumer app business. LTX-2.5 followed on August 11 — the vendor’s own blog describes Diffusion Fidelity Rendering, native multishot generation that holds continuity across cuts, a custom Gemma 4 text-encoder backbone, and full HDR ACES support. Within about ten hours it had open weights on Hugging Face, Day-0 native ComfyUI support, and a Runway integration. Licensing is unchanged in shape: free for organizations under $10M in annual revenue measured across the whole entity, paid above. By LTX’s own account the open-weights line has passed 22 million cumulative Hugging Face downloads since the base LTX-2 architecture launched in January 2026 — a vendor claim, not an independently audited figure.

LTX for Post-Production (September 29 – October 2, 2026). LTX’s release notes list six compositing, colour and resolution tools added in four days: Alpha Gen for alpha mattes, Layout to Render from 3D blocking, HDR Native (16-bit EXR in and out through ACES), an upgraded SDR-to-HDR conversion, a Native Resolution workflow that processes footage at delivery resolution on one GPU, and Refine and Restore, which LTX says combine with it to reach 8K. Tooling around a tracked model; none of it is scored on any board.

The launch claim and the board disagree, and the gap is large. LTX’s commissioned blind tests put LTX-2.5 at a 67% win rate, ahead of Seedance 2.5 (65%) and Gemini Omni Flash (55%) — figures reported by VentureBeat, which flags them as vendor-commissioned rather than independent. Artificial Analysis, blind-preference and independent, puts both LTX-2.5 tiers at Elo 1,060 on the audio text-to-video board, ranks 21–23 of 33 — roughly 180 Elo behind Omni Flash. On the no-audio board they read 1,213 (Fast) and 1,204 (Pro). Two weeks in, the Pro tier has not measurably beaten the Fast tier on any board — they are tied within their error intervals on text-to-video and Fast leads on audio image-to-video (1,044 to 1,013), while Pro lists higher. Treat the 67% as a marketing figure with a named methodology gap, not as a result (Weekly Roundup — August 17, 2026). (Retired v1 scale. On the rebuilt v2.0 board, October 11: Fast 946 and Pro 943 with audio, 938 and 934 silent.)

Wan 2.7 — Alibaba (Tongyi Lab)

Best for: Multi-task video generation with Thinking Mode through Alibaba Cloud’s API

Not to be confused with Wan 3.0, which Alibaba opened in invite-gated commercial beta on August 6, 2026 and is tracked in the commercial section. Corrected 2026-09-13: this entry said Wan 2.7 was the open-weights line. It is not. No Wan 2.7 weights exist on GitHub (Wan-Video tops out at Wan2.2) or Hugging Face (Wan-AI, the same), none were found on ModelScope, and Alibaba’s own launch coverage describes cloud and API access only. The open weights stop at Wan 2.2; Wan 2.7 sits in this section because it was placed here in April on a label nobody sourced, and it stays here pending a re-filing out of this section (the September 24 open-weights tracker records the finding; the move itself is open).

• Max resolution: 1080p
• Max duration: 2–15 seconds
• Task types: T2V, I2V, video continuation, reference-to-video (up to 5 persons), video editing
• Key feature: Thinking Mode (chain-of-thought reasoning before generation)
• Integration: Alibaba Cloud Model Studio, wan.video; ComfyUI 0.18.5+ (API nodes)
• License / weights: API and cloud only — no published weights. (Corrected 2026-09-13: this line read “Open source” from April 3 to September 13, 2026, with no vendor claim behind it.)
• Benchmark (July 2026): A dated checkpoint, Wan2.7-260612, debuted #2 on the Artificial Analysis text-to-video-with-audio board (Elo ~1,160, retired v1 scale; on T2V v2.0 Wan2.7-260612 is #13 with audio at 1,030) — the window’s only fresh top-5 leaderboard movement (Weekly Roundup — July 13, 2026)

Alibaba’s Wan 2.7, released April 3, 2026 (Weekly Roundup — April 17, 2026), is a major upgrade from the 2.2 line. The headline feature is Thinking Mode — a chain-of-thought reasoning approach where the model analyzes the prompt, plans composition, then generates. This produces noticeably more coherent output with fewer artifacts than single-pass generation.

Wan 2.7 Video unifies five task types in a single model: text-to-video, image-to-video (first-frame, first-and-last-frame, audio-driven), video continuation with text guidance, reference-to-video with up to five real-person inputs, and video editing via text, reference images, or style transfer. ComfyUI added support the same day in version 0.18.5 with workflow templates for all five task types.

HappyHorse-1.0 — Alibaba ATH AI Innovation Unit

Best for: Top-ranked benchmark quality (T2V + I2V); commercial API access with joint audio-video and seven-language native lip-sync

• Max resolution: 1080p
• Audio: Joint audio-video generation in a single forward pass; native synced output
• Lip-sync languages: 7 — English, Mandarin, Cantonese, Japanese, Korean, German, French
• Architecture: 15B-parameter unified 40-layer self-attention Transformer
• Inference speed: ~38 seconds for 1080p on a single NVIDIA H100
• Benchmark position (mid-July 2026, retired v1 scale): #2 on Artificial Analysis T2V no-audio (Elo 1,288), behind Gemini Omni Flash’s #1 (1,327); HappyHorse-1.1 #3 (Elo 1,273); Seedance 2.0 #4 (1,272). Omni Flash swept #1 on all four AA boards this month, ending HappyHorse-1.0’s no-audio reign. On the with-audio boards HappyHorse-1.1 sits #4 (T2V ~1,150) / HappyHorse-1.0 #5, both behind Omni Flash (#1) and Seedance 2.0 (#2)
• Access: API live via fal.ai ($0.14/sec 720p, $0.28/sec 1080p) and Alibaba Cloud Bailian (enterprise from April 27, 2026); open weights still pending despite ATH’s marketing claim
• API: ✓ via fal.ai (4 endpoints) and Alibaba Cloud Bailian (enterprise tier)

HappyHorse-1.0 debuted anonymously on Artificial Analysis on April 7, 2026 (Weekly Roundup — April 11, 2026), immediately ranked #1 in both text-to-video and image-to-video blind testing, surpassing Seedance 2.0. Alibaba revealed its ATH AI Innovation Unit ownership on April 10. The 15-billion-parameter model uses a unified 40-layer self-attention Transformer that generates audio and video jointly in a single forward pass — no cross-attention modules, no separate audio post-processing.

In April 2026, fal launched HappyHorse-1.0 as official API partner with four endpoints at $0.14 per second for 720p output and $0.28 per second for 1080p — pay-per-second, no minimums. Alibaba Cloud Bailian opened enterprise-grade access the same day.

The open-weights story is messier. ATH’s happyhorse.me/open-source landing page describes HappyHorse-1.0 as “fully open-sourced,” but independent verification finds a public GitHub repo with no model weights, no inference code, and no license file; the Hugging Face profile remains auth-gated. Alibaba has effectively separated commercial API access (live) from open-weight distribution (still unscheduled). Until weights ship, treat HappyHorse-1.0 as a commercial model with an open-source promise — the API is the actual access surface.

HappyHorse-1.1 appeared on the Artificial Analysis leaderboard in late June 2026 — the first model update from ATH’s AI Innovation Unit since the 1.0 debut in April — and has since settled into #2 on both Artificial Analysis with-audio boards: T2V with-audio (Elo 1,151, behind Seedance 2.0’s 1,222) and I2V with-audio (Elo 1,116, behind Seedance 2.0’s 1,194). HappyHorse-1.0 sits third and fifth respectively on those same boards (Elo 1,125 T2V; Elo 1,089 I2V) — Grok Imagine Video 1.5 and Wan 2.7 have both climbed above it on I2V since its debut. On the no-audio T2V board the ranking is unchanged: HappyHorse-1.0 #1 (1,290), HappyHorse-1.1 #2 (1,285). No separate pricing or distinct API endpoint has been announced for 1.1; access is through the same fal.ai and Alibaba Cloud Bailian surfaces. Open-weights delivery for 1.0 remains unshipped — 161 days past the April 27, 2026 commercial API launch and the “fully open-sourced” marketing claim. (The HappyHorse Elos in this entry are on the retired v1 scale. On T2V v2.0, October 11: HappyHorse-1.1 #12 with audio (1,042), HappyHorse-1.0 #14 (1,023); silent #10 (1,042) and #11 (1,038).)

MAGI-2 Preview — Sand.ai

Best for: Open-weights unified audio-video generation at the 100B+-parameter tier; Apache 2.0 with no territory exclusion, unlike MiniMax H3 and HY-World 2.0

• Architecture: 114B-parameter MoE (MagiMoE), ~6B parameters active per token; two-stage pipeline — magi2_preview (low-resolution denoising) + magi2_refiner (upscale to 1080p)
• Max resolution: 1080p (via refiner stage)
• Max duration: 10 seconds — the only duration currently supported
• Audio: Native — generated jointly with video in the same pass, muxed into the output file
• Checkpoint size: ~307GB total — 228GB preview-stage transformer, 56GB Qwen3.5-27B text encoder, 14GB refiner, 5GB Stable Audio Open 1.0 audio VAE, 3GB Wan2.2 video VAE, 2GB turbo VAE decoder
• Hardware: Eight NVIDIA Hopper-class GPUs (~80GB VRAM each) documented for inference; no smaller-footprint path published
• License: Apache 2.0 — unmodified boilerplate plus a copyright notice; no territory carve-out, community-license rider, or acceptable-use policy anywhere in the repository
• Access: Weights-only. The Hugging Face model card confirms it “isn’t deployed by any Inference Provider” — self-host or wait for third-party inference
• Arena (October 11, 2026): #10 on Artificial Analysis’s image-to-video leaderboard (Elo 1,093) and #21 on text-to-video v2.0 with audio (960; the board lists it as “MAGI-2 Preview (0912)”) and 957 silent (#20), a new scale on which its earlier 1,156 and 1,114 do not carry
• Vendor: Sand.ai (Beijing) — founded by Cao Yue, a Swin Transformer co-author and former Microsoft Research Asia researcher; prior release MAGI-1 (24B-parameter autoregressive video model, open-sourced April 2025)

Sand.ai released MAGI-2 Preview’s weights and inference code on August 5, 2026 — a 114B-parameter mixture-of-experts model that activates roughly 6B parameters per token to generate 10-second clips with synchronized audio in a single pipeline. The model card confirms the full spec: two-stage generation (a preview pass followed by a 1080p refiner), a ~307GB checkpoint across six weight sets, and a documented inference requirement of eight Hopper-class GPUs, with no smaller footprint published. It debuted at #6 on Artificial Analysis’s image-to-video leaderboard.

The license is the story. MAGI-2 Preview ships under a standard, unmodified Apache 2.0 grant — no separate community license, no territory clause, no acceptable-use policy layered on top. That’s a direct contrast with MiniMax H3, whose Community License withholds the US, EU, UK and South Korea from its grant entirely (MiniMax invites separate license requests from there). Sand.ai’s own about-us page calls MAGI-2 Preview “the world’s first 100-billion-parameter MoE video generation model” released as fully open source — a vendor claim, not independently audited, but the license terms behind it check out unmodified against the repository.

Sand.ai is a Beijing-based lab founded by Cao Yue, a co-author of the Swin Transformer who previously led a vision-model research center at the Beijing Academy of Artificial Intelligence. MAGI-1, the company’s first open-source video model, shipped in April 2025; MAGI-2 Preview is Sand.ai’s first appearance on this page.

Other Notable Open-Source Models

• SkyReels V4 (Skywork AI) — Released April 3, 2026 (Weekly Roundup — April 17, 2026). Presented at launch as the first open-source model to co-generate video and synchronized audio in a single forward pass. Dual-stream Multimodal Diffusion Transformer (MMDiT) architecture; 1080p at 32 FPS, clips up to 15 seconds. Accepts text, images, video clips, masks, and audio references. Ranked among the top models on Artificial Analysis T2V with audio leaderboard (Elo ~1,135 on the retired v1 scale; on T2V v2.0 it is #9 with audio at 1,074 and sixth silent at 1,084). The free tier was 70 monthly credits on skyreels.dev, now offline, and no weights have been published — Skywork’s GitHub org stops at SkyReels-V3 and a February 26 request for V4 weights on it sits unanswered. (Corrected 2026-09-13: this bullet said open-source weights were available for local deployment. They never were.)
• Kandinsky 6.0 Video (Sber’s Kandinsky Lab) — Open-sourced October 6, 2026 under the plain MIT License (copyright Kandinsky Lab, no territory, field-of-use or revenue condition), two days after its technical report. A joint audio-video diffusion family in two sizes, Lite (3B parameters) and Pro (29B), both making five-second clips with synchronized 44 kHz audio and lip-sync from text or an image; a separate MIT-licensed super-resolution model takes output to Full HD (1920×1080). Supported in Diffusers, ComfyUI, vLLM-Omni, SGLang and FastVideo. The README gives five-second-clip times on an RTX 4090 of 437 seconds (Lite, SD), 936 (Pro, SD) and 1,247 (Pro, Full HD), and says an 80 GB card needs CPU offload for Pro under vLLM-Omni. The main Pro checkpoint on Hugging Face sits behind a contact-information form while carrying the MIT tag; the Pro-distill and Lite repositories are ungated. Quality is untested: it is on no Artificial Analysis board as of October 11, 2026, and the paper’s own human evaluation says Pro outperforms Kandinsky 5.0 and is competitive with leading audio-video models, especially in speech quality (Weekly Roundup — October 12, 2026)
• Mochi 1 — High-fidelity short video with strong prompt alignment
• HunyuanVideo / HY-World 2.0 (Tencent) — HunyuanVideo offers solid image-to-video with coherent motion. In April 2026, Tencent’s Hunyuan team released HY-World 2.0 — a multi-modal world model that generates editable 3D scenes (meshes plus Gaussian Splattings) from text prompts or single reference images, with WorldMirror 2.0 inference code and weights open-sourced (github.com/Tencent-Hunyuan/HY-World-2.0) — under Tencent’s own HY-World 2.0 Community License, which excludes the EU, UK and South Korea and requires a separate licence above one million monthly active users (per the license text, September 13, 2026). The combination of editable 3D geometry and open weights makes HY-World 2.0 the more pipeline-friendly counterpart to Alibaba’s still-gated Happy Oyster
• Happy Oyster (Alibaba ATH) — Released April 16, 2026 (Weekly Roundup — April 17, 2026). World model that generates interactive, physics-aware 3D environments from text prompts; targets gaming, film, and VR. Directing and Wandering modes are designed for real-time exploration but don’t expose the underlying 3D representation in a standards-friendly way (unlike Tencent’s HY-World 2.0 above). Live demo accessible via Artificial Analysis arena; weights gated
• Helios (Peking University / ByteDance / Canva) — 14B autoregressive diffusion model; 19.5fps real-time generation on a single NVIDIA H100; capable of minute-scale video; Apache 2.0 license. Released March 2026. Notable for real-time throughput on a single accelerator
• NVIDIA Cosmos 3 (NVIDIA) — Released June 1, 2026 at Computex (Weekly Roundup — June 8, 2026). An open-weights omnimodel for physical AI that generates text, images, video, ambient sound, and actions in a unified architecture; shipped as Cosmos 3 Super and Cosmos 3 Nano on Hugging Face under the OpenMDW-1.1 license, corroborated by a 291-author arXiv paper. NVIDIA claimed top open-source rank on Artificial Analysis for text-to-image and image-to-video at launch; as of June 2026 the live board confirms Cosmos3-Super leading open-weight image-to-video at Elo 1,251 (Weekly Roundup — June 15, 2026) — the claim now borne out on the I2V side. Positioned as physical-AI / world-model infrastructure rather than a creator generation endpoint, but the open weights and unified video generation make it a tracked open-source entrant

How to Choose: A Routing Framework

The right model depends on the shot, not the project. Here’s a practical decision framework:

Which AI video model is best for broadcast-ready 4K? Kling 3.0 or Veo 3.1. Kling hits 4K at 60fps with multi-cut storyboards. Veo 3.1 leads on photorealism.

Which AI video model wins on benchmark quality with commercial API access? Wan 3.0 (#1 text-to-video v2.0 on both toggles; invite-gated, Alibaba Cloud Model Studio), or with a public price MiniMax H3 ($0.13/sec at 2K) and H3 Max via fal.

What’s the best free AI video model to start with? Veo 3.1 via Google Vids (10 free clips/month, any Google account).

Which AI video model is free inside an app you already use? Gemini Omni Flash via YouTube Shorts and YouTube Create (available since May 2026, any Google account).

Which AI video model accepts multimodal input (image + audio + video → video)? Gemini Omni.

Which AI video model wins character consistency across shots? Seedance 2.0 Pro via CapCut (US available since April 2026, with real-face restrictions) or Luma Ray 3.2.

Which AI video model is best for stylized and VFX work? Runway Gen-4 Turbo.

Which AI video model propagates a single-frame edit across a 30-second multishot? Runway Aleph 2.0 in Edit Studio.

Which AI video model handles professional production volume at scale? Luma Ray 3.2 (4× faster, 3× cheaper than previous Ray; adds 16-keyframe control and HDR/EXR export).

What’s the best low-cost AI video model for volume work? Pika 2.5.

What’s the cheapest AI video API in 2026? It depends on the quality you need. The board’s price floor is Agnes-Video-V2.0 at $0.30/min, near the bottom of the rankings (What’s Coming). Nearer the top, Wan 3.0 lists at $0.05/sec at 480p, H3 Max Turbo at $0.04/sec at 768P, and Grok Imagine Video 1.5 starts at $0.08/sec at 480p with native audio ($0.25 at 1080p).

Which AI video model is best for local generation and privacy? LTX-2.5 via ComfyUI or desktop editor.

Which AI video API is best for real-time interactive avatars? Runway Characters (GWM-1).

What’s the best real-time AI video for live agent meetings? PikaStream 1.0 (24fps/480p, ~1.5s latency).

Which AI video model wins multi-shot narrative? Seedance 2.0 Pro via CapCut (US, with restrictions), Luma Ray 3.2, or Kling 3.0 Omni.

Which AI video models work inside Adobe Creative Cloud? Adobe Firefly multi-model hub (Veo 3.1, Kling 3.0/Omni, Runway Gen-4.5, Luma, plus 30+ others).

Which AI video orchestration agent runs Kling + Veo + Seedance + MiniMax from one chat? Pika Agents (April 28, 2026; Slack/Telegram/Discord/X/Notion/Figma, persistent memory) or Higgsfield Supercomputer (mid-May 2026; orchestrates Seedance 2.0, Gemini, GPT-5.5 on web and Telegram). Runway Agent (May 13, 2026) covers the single-vendor end-to-end case. Luma Agents for enterprise campaign production (Publicis, Adidas, Mazda deployments). Adobe Firefly AI Assistant for teams already in Creative Cloud (public beta, CC subscription). See The Agentic Layer for the full comparison table.

Which AI model is best for editable 3D world generation? Tencent HY-World 2.0 (open weights) or Alibaba Happy Oyster (gated early access).

Which AI video model has the largest built-in distribution? Grok Imagine (500M+ X users).

Most professional workflows use 2-3 models per project, routing different shots to different engines based on the specific requirements of each scene.

What’s Coming

• Kling 4.0 — announced, not launched (as of October 11, 2026) — Kling’s release note says the full model launches in October and that Kling 4.0 Flash went to a limited group of users for early access on September 28. The note’s spec table lists 3–30 second clips, 720p / 1080p / 4K, 21:9, two-channel stereo audio, up to 10 keyframes, Omni Reference with up to 15 items across images, videos, voice references and subjects, 8,000-token prompts and 10-bit HDR at 1080p and 4K; the body of the same note puts HDR output and video extension to two minutes under ‘coming soon’. Flash is narrower: 3–20 seconds, 720p, 8-bit SDR, text-to-video, first-frame image-to-video and Omni Reference. Kling’s API pricing page says its packages currently don’t support Kling 4.0 and has no 4.0 or Flash row; the homepage leads with ‘All-New Kling 4.0’ and a Create Now button, and Kling’s October 9 post refers to ’the launch of Kling 4.0’ ahead of an October 14 panel in Santa Monica. Kling’s own note and price page say launch is still ahead. The specs are Kling’s alone and untested; Kling 4.0 is on no Artificial Analysis board (the newest Kling entries are 3.0 Omni, 1,015 with audio, and 3.0, the 1,000 anchor, on the text-to-video v2.0 board). Price is unpublished (Weekly Roundup — October 12, 2026)
• Reactor — real-time world models as infrastructure — Reactor emerged from stealth in May 2026 with $59M (Lightspeed led; Jeffrey Katzenberg’s WndrCo, Amplify Partners, Sky9 Capital, FPV Ventures also in). Co-founders are former Apple Vision Pro technical leads. The pitch: a unified SDK and API that makes real-time world models available to developers “with a few lines of code,” targeting media and entertainment, physical AI, and robotics. This is a different category from the commercial models and agents tracked above — it is infrastructure for interactive AI worlds, not a generation endpoint. No Stack row warranted yet; worth watching as the definition of “AI video” expands toward interactive, real-time, and physics-driven output.
• MCP as a distribution layer for AI video — In May 2026, Runway launched Runway MCP, connecting its model roster (Gen-4.5, Seedance 2.0, GPT Images 2.0, Kling) to Claude, ChatGPT, Cursor, Replit, and any MCP-compatible client. The following day, Pika and Higgsfield both shipped their own MCP skills — Pika’s Founder Starter Kit (four Claude skills: Build-a-Brand, App Screens, Product Sizzle, Founder Video) and Higgsfield Supercomputer as a Claude skill. Three Tier-1 AI video labs shipped MCP integrations inside 24 hours; Higgsfield also launched five Adobe Premiere Pro and After Effects plugins in the same window. MCP is becoming the standard distribution channel for AI video capability into developer and coding-agent workflows — a second distribution layer running alongside the model APIs. In September 2026 Runway listed its MCP on Cursor’s Marketplace, let MCP agents read Brand Kits and shipped a plugin into Grok Bot, and Pika opened its API to Grok Bots.
• Seedance 2.5 — three aggregators shipped it the same Friday morning, still no price on any of them — ByteDance previewed Seedance 2.5 at Volcano Engine FORCE (June 23, 2026): single clips up to 30 seconds with no post-stitching, joint audio-video generation, up to 50 multimodal reference inputs, targeted post-generation editing. ByteDance’s own Dreamina app pushed the global consumer launch July 31; the BytePlus ModelArk developer API opened July 16. On August 7, 2026, Runway, Higgsfield, and Pika all went live with Seedance 2.5 within about two hours of each other (06:11–07:59 UTC) — the aggregator layer racing a model none of them controlled, resolving within a single morning. No independent benchmarks exist yet — every spec is ByteDance’s own claim. If durability holds, 30-second native generation from a top-tier T2V model resets the category’s duration ceiling. August 2026 additions: ComfyUI shipped native Seedance 2.5 partner-node support on August 7, putting the model on the self-hosted local-inference layer with a documented per-run cap of 50 reference assets (30 images, 10 videos, 10 audio), second-level shot control via timeline prompting, and native dialogue and lip sync in 10+ languages. On August 14 Higgsfield opened free 1080p generation as a “limited time” early-access tier — the first resolution figure attached to Seedance 2.5 on any surface, since the August 7 go-live announcements specified none. Pricing landed — corrected 2026-09-01. This page said “no published consumer or developer price from any aggregator” through August. That was wrong: ByteDance publishes first-party rates on BytePlus ModelArk for dreamina-seedance-2-5-260628 — $0.514 per 5-second 480p clip ($0.103/sec), $1.156 at 720p ($0.231/sec), $2.843 at 1080p ($0.569/sec), verified against the vendor’s own table. A dated 28% discount applies to 1080p only (480p and 720p explicitly excluded), running 14:00 UTC+8 August 14 through 14:00 UTC+8 September 17, 2026 — which puts the currently-billed 1080p rate near $0.41/sec, not $0.569. The list figure is computed off the pre-discount token rate. Anyone budgeting before mid-September should use the effective rate (Weekly Roundup — August 17, 2026). (Superseded 2026-09-27: the discount window closed September 17 and the BytePlus page, fetched September 27, no longer lists it; 1080p bills at the $2.843 list figure, $0.569/sec.) Draft tiers, September 23–24: the same BytePlus page documents a two-step Seedance 2.5 Draft mode — the draft billed as a 480p video, the final billed at its target resolution. Three storefronts surfaced it within about 41 hours: Pika (Draft Mode, September 23), Higgsfield (Preview Mode, 480p draft to 1080p render, September 23) and Runway (Draft mode, September 24) — one ByteDance feature, three launch posts.
• Meta Muse Video — MSL enters generative video — Meta Superintelligence Labs previewed Muse Video on July 7, 2026, alongside Muse Image — its first in-house image and video models after years of licensing Midjourney and Black Forest Labs. Built on the Muse Image pretraining base with native audio; Meta claims competitive prompt adherence, visual fidelity, and temporal consistency, and it debuted #3 on Design Arena’s text-to-video board (Elo 1,459, July 5). But it’s a preview, not a product — no published resolution, clip length, or pricing (“coming soon to creators and in Meta AI”), and Meta flags its own gaps in audio-video sync and fast motion. A watch entrant, not a counted Big Ten row until it ships with specs (Weekly Roundup — July 13, 2026); on the Design Arena board dated September 21, 2026 Muse Video sits #9 (Elo 1,456). Muse Realtime Avatar (September 23, 2026) is the second video model in the family and the first with named competitors: Meta’s post describes an audio-driven diffusion transformer that turns Muse Realtime Voice into an interactive avatar from reference media, streaming 448×768 portrait video at 25fps with about 870ms from the end of a user’s turn to first byte, after self-forcing and distribution matching distillation cut each chunk from 120 function evaluations to 2; 12 concurrent sessions run on one GB200. In Meta’s own rater study against Runway Characters and HeyGen LiveAvatar, raters preferred Muse overall and on every dimension, with the mannerism comparison against Runway Characters a statistical tie — Meta’s study, no outside replication. Output carries a Meta Video Seal watermark (Weekly Roundup — September 28, 2026)
• PixVerse — funded to scale, world model incoming — the Singapore lab (founded 2023 by ex-ByteDance vision lead Wang Changhu) closed a $439M Series C extension on July 13, 2026, pushing its valuation past $2B with Alibaba among the new backers (and already a deployment partner). PixVerse claims 150M+ users across 177 countries; the round funds its R1 real-time world model and a push into interactive entertainment. PixVerse V6 currently sits mid-table on the Artificial Analysis audio board (Elo ~1,070 on the retired v1 scale; #18 on T2V v2.0 at 985 with audio, #15 at 996 silent) — a watch entrant with money and distribution, not a counted Big Ten row until a category-leading model ships (Weekly Roundup — July 20, 2026)
• EU AI Act transparency Code of Practice — initial signatory list published July 31, 2026 — The EU AI Office published its marking-and-labelling Code of Practice on June 10, 2026, ahead of the binding Article 50 disclosure date. Google formally signed July 24; Meta followed July 28, framing the move as keeping disclosure “practical, interoperable and genuinely useful” while warning against “a growing array of different labels and disclosures that end up overwhelming people.” The Commission’s own published list (about 190 organisations; 95 signed the provider section) confirms it: of the labs this page tracks, Google, Meta, Black Forest Labs and OpenAI signed the provider section (so did Synthesia). ByteDance, Kuaishou, Alibaba, TikTok, SpaceXAI and MiniMax are not on the list as read October 11, 2026; the code remains open for signature, so absence isn’t necessarily a declined invitation. (Corrected 2026-10-11: this item said Google and Meta were the only AI-video-relevant signatories. The Commission’s text of July 31 names Black Forest Labs among Section 1 signatories.) (Weekly Roundup — August 3, 2026)
• Dreamina Octo — ByteDance’s “next chapter” beyond Seedance 2.0. Revealed at AI on the Lot (Culver City, May 27, 2026) under the framing “From Generation to Emergence” and “when the prompt isn’t the point.” Early access survey live; product confirmed as “arriving soon,” not yet shipped. If Octo ships as a conversational orchestrator rather than a generation model, it belongs in the agentic section; if it ships as a new Seedance-tier model, it belongs in the Big Ten. Watch @dreamina_ai for the actual launch.
• Gemini Omni Pro shipping window — Omni Flash’s June 30, 2026 public-preview API launch answered the access half of this item; Omni Pro itself remains unshipped with no date attached — “soon” is still the operative word. Omni Flash landed #1 on both Artificial Analysis with-audio boards July 14, 2026 (July, retired v1 scale for text-to-video: T2V 1,240 / I2V 1,203), passing Seedance, on top of its July 2 #1 debut on Design Arena’s Video Arena (Elo 1,404) — two boards, two methodologies, both now led by Omni Flash (RCTV flagship analysis →)
• Agentic-orchestration layer — Six conversational agents now sit on top of the model layer: Luma Agents (March 5, 2026), Pika Agents (April 28, 2026), Runway Agent (May 13, 2026), Higgsfield Supercomputer (mid-May 2026), Adobe Firefly AI Assistant (public beta April 27, 2026), and HeyGen HyperFrames (June 26, 2026). See The Agentic Layer for the full comparison
• Grok Imagine 1.5 at 1080p — resolved July 31, 2026 — native 1080p for text-to-video and image-to-video shipped July 31, 2026 per SpaceXAI release notes, as a priced tier ($0.25/sec, per xAI’s model page). Detail in the entry
• HappyHorse-1.0 open-source weights — Now 161 days past the April 27, 2026 commercial API launch on fal, and the weights still haven’t shipped: the public GitHub repo remains empty (no weights, no inference code, no license file) and the Hugging Face profile still reads “coming soon.” Artificial Analysis now lists HappyHorse as an API-only product. ATH’s happyhorse.me/open-source “fully open-sourced” claim hasn’t softened, hasn’t acknowledged the gap, and hasn’t put a date on the artifact. Independent verification by WaveSpeedAI remains the canonical source on the marketing-vs-artifact gap
• NVIDIA to acquire Hugging Face — the shelf every open video checkpoint sits on (September 3, 2026) — NVIDIA’s own announcement puts the price at $12.93 billion. MiniMax H3, MAGI-2 Preview, LTX-2.5, Wan 2.2 and NVIDIA’s own Cosmos builds all distribute from Hugging Face; after this, the neutral distribution layer for open video weights has an owner with a model line of its own. Carried for the distribution-terms question, not as a video story; watch what happens to hosting terms for tracked open-weight models once the deal closes (Weekly Roundup — September 7, 2026)
• TAKE IT DOWN Act enforcement first action — The FTC opened formal enforcement after the May 19, 2026 deadline (FTC press release, business-guidance framework). The agency stood up TakeItDown.ftc.gov as the victim-facing intake surface and on the same day sent a second-wave warning letter to twelve “nudify” tool sites — on top of the fifteen platforms named May 13, 2026. First enforcement target — platform, tool, or generator — sets the operative precedent for the rest of the year. Civil penalties up to $53,088 per violation; 48-hour removal SLA; Section 5 enforcement (Section 230 is no shield)
• AI executive order — signed June 2, 2026, light-touch — Trump signed Promoting Advanced Artificial Intelligence Innovation and Security on June 2, 2026. The version that landed is voluntary, not the FDA-style framework early drafts floated: developers of “covered frontier models” may give federal agencies a 30-day pre-release look, and the text explicitly bars any “mandatory governmental licensing, preclearance, or permitting requirement.” OpenAI, Anthropic, and Google all welcomed it. For AI video the exposure runs through the frontier-model layer, and the DOJ criminal-AI priority stacks onto TAKE IT DOWN. The month-long regulatory overhang resolved in industry’s favor (Weekly Roundup — June 8, 2026)
• Runway Cosmos Coalition — world models as an industry alliance — In June 2026, Runway, NVIDIA, and “leading AI labs” announced the Cosmos Coalition to build and open-source frontier world models; Runway joins as a founding member. As announced it’s a mission statement — no specs, no timeline, no architecture. It landed the same day as NVIDIA’s actual Cosmos 3 release (now in Open-Source above) and Luma’s Open Physical AI Lab, making “world model” three different things in 24 hours (Weekly Roundup — June 8, 2026). Runway’s own world-model work has since split in two: Solaris (August 31, 2026) generates interfaces and GWM Worlds 2 (September 3) generates video, both request-form previews — see the Runway entry
• Atlas — World Labs (September 1, 2026): a world model that takes the camera as an input — Atlas is pretrained from scratch on text, images, video and 3D, accepts camera poses natively, and generates up to a minute of video at 1440p along a supplied path; it also reconstructs 3D from as few as two or three images and will power future versions of Marble. World Labs’ own head-to-head chart (raters preferring Atlas’s camera-following over MiniMax H3 75%, Gemini Omni Flash 81%, Seedance 2.5 94%) comes with the vendor’s own disclosure that the rivals got the camera path as text — a measure of prompt-described camera control, not a quality ranking. Early access with select partners, no price, no API date, no independent evaluation. Tracked here, not counted in the Big Ten, until something is routable (Weekly Roundup — September 7, 2026)
• HiDream-O1-Video — a #4 image-to-video debut with no announcement behind it (September 2026) — HiDream.ai (Beijing) entered Artificial Analysis’s image-to-video board at #4, Elo 1,186, between September 9 and 10, above Gemini Omni Flash and Wan 3.0, sat #4 at Elo 1,176 on September 20, and #6 at 1,175 on October 11, in a five-way statistical tie for second through sixth with H3, Vidu Q4 Preview, Omni Flash and Seedance 2.0 720p (AA ranges 2–6). The board tags it “Released Aug 2026”; the vendor’s own site shows an O1 image model, an O1 world model and an O1 embodied model, and no page for this one — a model that reached a top-five board slot before its maker announced it. A watch entrant, not a counted row, until HiDream says what it is (Weekly Roundup — September 14, 2026)
• Bach 1.0 Pro — Video Rebirth (Singapore): a top-three no-audio image-to-video entrant with no Pro tier on the vendor’s own site (September 2026) — Artificial Analysis’s image-to-video board lists Bach 1.0 Pro third on its no-audio ranking at Elo 1,359 as of September 20, behind Gemini Omni Flash (1,369) and Wan 3.0 (1,361) and ahead of MiniMax H3 (1,354); it does not appear in the with-audio top ten. The vendor is real and traceable: Video Rebirth Ltd, headquartered in Singapore, founded by Dr. Wei Liu (formerly a Tencent distinguished scientist), $80 million raised in total, launched Bach on May 7, 2026 as a native-1080p, 30-second multi-shot engine, per its own release. Neither the release nor videorebirth.com names a “Pro” tier, so the Pro specs, price and ship date are unverified and not carried. A watch entrant, not a counted row, and a vendor added to tracking from a board this page reads but had only been reading on one of its two toggles (Weekly Roundup — September 21, 2026) As of October 11, 2026 it ranks second on the no-audio set at 1,362, behind Omni Flash (1,368) and level with Wan 3.0 (also 1,362), and the board dates it September 29 at $7.20 per minute — Artificial Analysis’s figures, still without a Pro tier on the vendor’s site. On September 24 Video Rebirth also open-sourced HyperFlow, an 8-step LoRA for MiniMax H3 (see H3).
• Vidu Q4 Preview — a provisional #3 image-to-video debut, with the headline price absent from its own price table (October 7, 2026) — Vidu released Q4 Preview as its ’next-gen flagship’. Artificial Analysis scores it third with audio on the image-to-video v1.0 board at Elo 1,179 (±10, labelled provisional), a leap from Vidu Q3 Pro’s 19th place (1,056); ranks two through six overlap, so it is a tie, not a podium. It does image-to-video and reference-to-video only, per Vidu: up to 15 reference images, three voice references, 3–16 second clips, up to 4K; the API model ID is viduq4-preview and AA says the full release is still ahead. Vidu’s API price table lists 9 / 19 / 24 / 38 / 78 credits per second at 540p / 720p / 1080p / 2K / 4K (1 credit = RMB 0.03125), the same 1080p rate as Q3 Pro. The launch post’s ‘from $0.014/sec’ is described by Vidu’s own reply as launch promotional pricing and does not appear on the table. A watch entrant with a public API, not a counted Big Ten row (Weekly Roundup — October 12, 2026)
• Agnes-Video — a new API price floor — Singapore’s Sapiens AI entered the Artificial Analysis arena with Agnes-Video-V2.0 at $0.30/min — the cheapest price on the board, roughly a tenth of the next commercial tier. The catch is quality: it debuts near the bottom (Elo 905, #24 on audio T2V). A price-floor entrant, not a quality threat — yet (Weekly Roundup — June 8, 2026) Update, September 13, 2026: Sapiens now lists a second tier, Agnes-Video-2.5, at #19 on text-to-video (Elo 1,085) and $1.50/min — five times V2.0’s price for roughly 170 Elo, while V2.0 holds the board’s last place (#36). Still a price-floor entrant, now with a mid-table tier above it. Update, October 11, 2026: on text-to-video v2.0 Agnes-Video-2.5 is #10 with audio (1,057) at $2.40/min per Artificial Analysis; Agnes-Video-V2.0 is not on that board (it remains on image-to-video v1.0 at 924)
• Varya — India’s government-backed open-weight entrant — Bangalore’s Avataar.ai launched Varya (June 11, 2026), an open-weight video model distilled from Alibaba’s Wan 2.2 (4 steps vs. 50; 5-second 720p in 45s on an H200), weights on India’s AI Kosh repository under the $1.2B IndiaAI Mission. The hook is price: a vendor-reported ₹0.48/sec (~$0.005), roughly 20× below the global leaders. Treated as a watch entrant, not a tracked open-source row: the 14B parameter figure is Indian-outlet-only (not confirmed by TechCrunch), Artificial Analysis hasn’t benchmarked it, and Avataar’s own model page was unreachable at writing. Real as a price-floor and sovereign-AI signal; specs unverified (Weekly Roundup — June 15, 2026)
• Black Forest Labs ships the video model — FLUX 3, July 23, 2026 — the pivot Martin Scorsese’s June 2 advisory role implied was coming: a multimodal model that “jointly learns from images, videos, and audio within a unified architecture,” generating up to 20 seconds per clip with native synced audio (text-to-video, image-to-video, video-to-video, multi-shot chaining). Early Access only via API/private weights; an open-weight “FLUX 3 Dev” is named with no ship date beyond “the next few weeks and months.” Runway and Pika both integrated it August 4 (Pika’s post landed about an hour before Runway’s — corrected 2026-09-06 from “Pika followed August 5”), Luma made it three on August 12, and fal.ai was serving it by August 26 at $0.17/sec (720p) and $0.29/sec (1080p) — four aggregators in about three weeks, the same fast-absorption pattern already established for Seedance 2.5, Grok Imagine, and MiniMax H3 (Weekly Roundup — August 17, 2026); Luma added FLUX Upscale on September 1. FLUX 3 Video debuted at #2 on Design Arena’s text-to-video board (Elo 1,496) on August 11 and sits at #3 (Elo 1,493, still marked “Preliminary”) on the board dated September 21, 2026; Artificial Analysis’s rebuilt text-to-video v2.0 board (September 30) lists FLUX 3 — #6 with audio (1,128) and #5 silent (1,091), ahead of Seedance 2.0 and Omni Flash 1.1 on both. Still primarily an image-model company by revenue and history; still a watch entrant, not a counted Big Ten row — verdict recorded September 10, 2026. The three-aggregator threshold was reviewed on August 18, 2026 and read as a pattern rather than noise; RCTV’s analysis of whether that makes BFL a commercial frontier lab or the infrastructure underneath the platforms absorbing it found the marketing and the money pointing in different directions, with only the infrastructure reading costing anything to maintain: utility pricing on the one capability shipped since launch, 21 days of silence on @bfl_ai by publish day, and FLUX listed under image partners only on Adobe’s own partner page while three of BFL’s video aggregators headline Adobe’s video section. Big Ten membership reopens on any of four signals the piece names — a fifth aggregator, an enterprise or funding disclosure, an Artificial Analysis board entry for FLUX 3 Video, or @bfl_ai posting again. First capability update since launch: FLUX Upscale (August 20, 2026), a dedicated native-resolution upscaler taking FLUX 3 Video output to 2K/4K, via API and a public demo — shipped after @bfl_ai had gone four consecutive pulse runs silent. A lab shipping pipeline utilities for its own model was the evidence the analysis leaned on hardest. Then, one day after the analysis published, @bfl_ai broke its silence: FLUX Video Edit (September 10, 2026) — prompt-driven edits to an existing clip across eight edit types (add, remove or replace objects and characters, restyle, rebuild settings, change or translate dialogue with lip sync, and more), $0.03 per second of output, about 50 seconds for a 10-second clip, live on fal.ai the same day. That is one of the four reopen signals firing — and an Artificial Analysis board entry for FLUX 3 Video, a second, has now fired too — the question is reopened, not re-answered — and the shape of the ship, a fast and cheap utility, is the shape the analysis read as infrastructure. Runway widened its BFL shelf on September 16: its “Fall 2026 collection” added Flux Video Upscale and Flux Video Edit — an existing aggregator carrying more of BFL’s line, not the fifth aggregator the reopen test names (Weekly Roundup — September 21, 2026). FLUX 3 Action (September 23, 2026) is an open-weights 7B World Action Model for robot control, built on FLUX 3’s image, video and audio pretraining with a smaller architecture — robotics, not video generation, and nothing the reopen test counts
• State deepfake legislation map — Federal preemption of state AI law isn’t happening in 2026. Connecticut HB 5312 (May 2026) establishes a private right of action against creators of non-consensual AI-generated intimate imagery. Vermont’s election-deepfake bill, Iowa’s chatbot-safety law, and Utah’s nine AI bills moved in the same period. Together with Tennessee’s ELVIS Act, California’s AB 2655, and New York’s election-deepfake law, the state map is denser than the federal one. AI video labs allowing image-to-video from real-face inputs need to model state civil exposure alongside federal compliance
• State synthetic-media provenance mandates — A second state-law track, distinct from the NCII-liability bills above: provenance-embedding requirements that reach the generators directly rather than punishing misuse. Connecticut SB 5, the omnibus AI act, is law as Public Act 26-15: the General Assembly’s bill history records it signed by the Governor on May 27, 2026 (a law-firm summary says June 2). Its provenance duty reaches providers of consumer-facing generative AI systems with more than 1M monthly users, who must use commercially reasonable measures, such as C2PA, to make provenance data in generated audio, image and video hard to remove or alter; per Kilpatrick Townsend’s summary it began October 1, 2026, and the Attorney General enforces it. We have not read the Act’s text for these dates, and the October 1, 2027 detectability standard this page carried is unconfirmed, so it is removed. Arizona SB 1786 died — recalled to the House for reconsideration (the Senate returned it May 4, 2026) and never enacted; the azleg.gov Bill Status Inquiry lists it “Engrossed — Dead,” confirming RCTV’s R#12 correction that it died in the legislature and was not vetoed. California’s AI Transparency Act — SB 942 (2023–2024, Chapter 291), as amended by AB 853 and, since September 30, 2026, by SB 1000 — is the operative California statute, in force since August 2, 2026. SB 1000 was approved September 30 as Chapter 861, Statutes of 2026, an urgency statute that takes effect immediately, per the chaptered text. It deletes the 1,000,000-monthly-user threshold from the definition of “covered provider,” renames the “AI detection tool” a “disclosure verification tool,” drops the user-facing manifest-disclosure option, requires the latent disclosure to state whether the system created or altered the content, and exempts systems designed primarily as assistive technology until January 1, 2029, with a $50,000-per-violation penalty for falsely claiming the exemption; the general penalty is $5,000 per violation (Business and Professions Code § 22757.4). The disclosure verification tool (§ 22757.2) must be free and publicly accessible, accept an upload or a URL, output any system provenance data it detects, and support technology, ‘including an application programming interface,’ so a user can invoke it without visiting the provider’s website; a provider may impose reasonable access limits to prevent or respond to security risks or misuse, and may satisfy the duty by directing users to a third-party tool that itself complies, is compatible with the provider’s latent disclosures and is clearly accessible from the provider’s interface. Google’s SynthID Detector (public October 7) bars programmatic submission in its terms; neither Google’s announcement nor the portal mentions California, and nothing we read says whether the Detector is the tool the statute contemplates (Weekly Roundup — October 12, 2026). AB 2713, approved the same day as Chapter 856, bars a “large online platform” — more than 2,000,000 unique monthly users — from knowingly stripping system provenance data or digital signatures where technically feasible, operative January 1, 2027. SB 1111, Chapter 862, adds a “digital replica” to the voice-or-likeness definition in Civil Code § 3344 and makes using one with intent to impersonate false impersonation under a new Penal Code § 540 (Weekly Roundup — October 5, 2026). Earlier, SB 1050, synthetic-performer disclosure in advertisements — the state’s counterpart to New York’s law below — cleared the legislature August 31 (Assembly concurrence 40–0), was presented to the Governor September 9 and was signed and chaptered September 16, 2026 as Chapter 246, Statutes of 2026, verified against the leginfo record. It adds Business and Professions Code § 17610: an ad that prominently features a synthetic performer — defined as a generative-AI figure that reads as a human performance while being “not recognizable as any identifiable natural person” — must carry a clear and conspicuous disclosure, with wording substantially like “this performance features a synthetic performer”; ads for expressive works are exempt when the synthetic performer is used as it is in the work. No urgency clause, so it takes effect January 1, 2027 on the standard schedule (Weekly Roundup — September 21, 2026). SB 1000 amends the regime the August 24 roundup covered when an audit found most SB 942-covered companies without a working detector, as the Assembly Privacy and Consumer Protection analysis anticipated. Hawaii HB 2137 was signed — it became Act 247 on July 14, 2026 (Gov. Msg. No. 1349), “Relating to Artificial Intelligence: Realistic Digital Imitations; Protections for Individuals,” after passing both chambers on final reading May 6 and reaching the Governor May 7. Verified against the Hawaii Legislature’s own measure-status record. Unlike removal mandates, provenance embedding is a model-build requirement — it changes what the model ships, not just what a platform takes down (Weekly Roundup — May 25, 2026)
• The training-data cases — live, and deliberately carrying no trial date — Andersen v. Stability AI (N.D. Cal., 3:23-cv-00201, Judge Orrick, filed January 2023) is the generative-media fair-use case whose outcome would govern training-data law for video models as much as image ones, which is why it sits on a page about video. RCTV is not publishing a trial date for it. Trade coverage has circulated one; the court’s own public case page carries no trial-date field, govinfo’s case record carries no scheduling data, and the docket’s own recent-filings list, read September 27, 2026, runs through September 25 on the same contested ground: the artist plaintiffs moved to modify the case schedule on September 21 (the requested changes are redacted), Stability AI opposed on September 25, and a sealed sanctions reply against DeviantArt landed September 22. Until the court rules on that motion, any trial date in circulation is a guess, and this page would rather print nothing than a date sourced to repetition. Separately and firmly on the record: Stability AI closed a $76M Series B on August 25, 2026 — $232M total to date — with Electronic Arts and all three major music groups (Sony Music, Universal Music, Warner Music) taking equity, alongside AMD Ventures and Pacific Alliance Ventures. Rightsholders buying into the company they are litigating against is the more interesting fact than any scheduling order
• Hollywood and ByteDance sign — the first AI-safeguards accord between the MPA and a Chinese tech company (August 17, 2026) — ByteDance and the Motion Picture Association signed a global memorandum of understanding on AI intellectual-property protections covering Seedance (video) and Seedream (image) across TikTok, the TikTok USDS joint venture, CapCut and Dreamina, per the MPA’s own announcement. Read what it is: a voluntary truce, not a licensing deal — studios receive no payment for catalogue use, and neither party published technical enforcement details, thresholds, or detection mechanisms alongside it. It resolves the enforcement chain running from the February 12 cease-and-desist over Seedream 5.0 Lite and Seedance 2.0 through the March 9 legal escalation (Weekly Roundup — August 24, 2026)
• New York Synthetic Performer Disclosure Law — live June 9, 2026 — First-in-the-nation requirement that advertisers conspicuously disclose AI-generated “synthetic performers” (fabricated human likenesses) in visual/audiovisual ads reaching New York consumers; civil penalties $1,000 first violation / $5,000 subsequent; audio-only ads and expressive works (film, TV, games) exempt. A distinct mechanism from content-labeling — ad-specific disclosure — adding a jurisdiction to the compliance staircase (Weekly Roundup — July 13, 2026)
• Runway Gen-4.5 — Accessible via Adobe Firefly’s multi-model hub since April 2026; standalone Gen-4.5 launch on runwayml.com still pending. Previewed on NVIDIA Vera Rubin hardware at GTC (March 2026)
• NVIDIA Vera Rubin cloud deployment — AWS, Google Cloud, Microsoft Azure, and OCI all confirmed H2 2026 availability. Vera Rubin delivers 10× lower inference token cost versus Blackwell — the number that will reshape per-second AI video pricing across all major cloud platforms
• DLSS 5 — NVIDIA’s neural rendering technology, launching Fall 2026. Explicitly positioned for filmmaking and VFX beyond gaming; uses generative AI to infuse photoreal lighting and materials anchored to source 3D geometry
• Blackburn draft AI bill — GOP Senate draft (March 19, 2026) declares AI training on copyrighted works not fair use; targets deepfakes and Section 230. Not yet introduced as legislation; path to passage uncertain
• White House AI framework vs. CLEAR Act — White House (March 2026) takes the opposite position from Blackburn: AI training is not infringement; courts should decide. Bipartisan CLEAR Act (Schiff/Curtis) proposes mandatory training data disclosure without resolving fair use. Three irreconcilable positions now active in Washington simultaneously
• Seedance 2.0 copyright litigation — US CapCut access available since April 2026 with real-face and IP restrictions, but the underlying copyright dispute with Disney, Paramount, Warner Bros., and Netflix remains unresolved. The restrictions are a negotiating posture, not a settlement
• OpenAI robotics / world simulation — OpenAI redirected Sora’s compute toward “world simulation for robotics” after shutting the product down. The consumer app went dark on April 26, 2026 as scheduled; the Sora API — every model alias and snapshot — was removed September 24, 2026 with no replacement listed on OpenAI’s deprecations page, which ends OpenAI’s presence in video generation entirely (Weekly Roundup — September 21, 2026)
• Adobe Firefly multi-model expansion — Firefly’s video hub now hosts 30+ third-party AI models including Kling 3.0/Omni, Veo 3.1, Runway Gen-4.5, ElevenLabs Multilingual v2, Luma AI, Black Forest Labs, and Topaz Labs. Firefly AI Assistant orchestrates multi-step workflows across Photoshop, Premiere, Lightroom, Express, and Illustrator
• Tencent vs. Alibaba 3D world model race — Two of China’s largest AI labs shipped 3D world models on the same day, April 16, 2026 (Alibaba’s Happy Oyster, gated; Tencent’s HY-World 2.0, open weights). Western labs have nothing comparable in production; the 6-to-12 month head start is real if world simulation matters as much as OpenAI’s Sora-shutdown framing implied
• Google Vids / Workspace expansion — YouTube export is live; paid creative tiers (Pro/Ultra) include Lyria 3 music generation and AI avatars. Further Workspace AI integration expected throughout 2026
• EU AI Act Article 50 — in effect since August 2, 2026, with one grace period. Article 50’s disclosure and deepfake-labelling duties applied on August 2, 2026 under Article 113 of the Act. Fines for breaching the Article 50 transparency duties fall under Article 99(4)(g): up to €15 million or 3% of worldwide annual turnover, whichever is higher. The harder machine-readable marking duty (Art. 50(2)) carries a grace period to December 2, 2026 for providers of generative systems placed on the market before August 2, 2026, under Article 111(4), added by Regulation (EU) 2026/1744 (Weekly Roundup — August 3, 2026)
• EU AI Act Digital Omnibus — Regulation (EU) 2026/1744: new prohibitions on non-consensual intimate imagery and CSAM generation apply from December 2, 2026 — The Omnibus was published July 24, 2026 and entered into force July 27. It inserts Article 5(1)(ba) and (bb), prohibiting AI systems that generate or manipulate realistic imagery of an identifiable person’s intimate parts or sexually explicit activity without that person’s explicit consent, and material within the meaning of Directive 2011/93/EU (child sexual abuse), except where a ‘without right’ defence applies under national law. For providers the ban applies where the generation is the system’s intended purpose, or where its design, training, architecture, capabilities or user-facing features make it a reasonably foreseeable and reproducible outcome and it lacks adequate technical safeguards; deployers are caught when they use a system for that purpose (Art. 5(1a)). Article 5 breaches carry fines of up to €35 million or 7% of worldwide turnover, whichever is higher (Art. 99(3)). The same act sets December 2, 2026 as the end of the Article 50(2) marking grace for systems placed on the market before August 2 (Art. 111(4)) and moves the high-risk rules to December 2, 2027 (Annex III systems) and August 2, 2028 (Annex I systems). It leaves Article 50’s own application date, August 2, 2026, unchanged.
• Unlimited-length AI video — EPFL’s drift elimination breakthrough (presenting at ICLR 2026) could remove the duration ceiling entirely
• SpaceXAI targeting 30-minute video — Announced goal for late 2026, with full-length films targeted for 2027

This page is maintained by RCTV as a public reference. For weekly updates on model releases and industry shifts, see our Weekly Roundup.

Have a correction or update? Contact us at rctv.oxncw@simplelogin.com

Changelog

Showing the four most recent updates. Full changelog archive →

October 11, 2026

• Paired refresh with the October 12 Weekly Roundup: lastmod → 2026-10-11 (lead-by-one before the October 12 Roundup). The “Last updated” line now reads October 11 and the Quick Verdict “current as of” and Quick Reference caption dates move with it.
• CORRECTION — Grok Imagine Video 1.5 is priced by resolution, and this page said it had one rate: xAI’s model page lists $0.08 / $0.14 / $0.25 per second at 480p / 720p / 1080p; the $0.08 figure is the 480p rate and a 1080p minute lists at $15.00, not $4.80. An archive capture of July 18 shows the same table, so the “single published rate, no resolution tier” line was wrong when the October 4 pass restated it. The intro, the Quick Verdict “Simplest published API price” bullet (replaced by a resolution-tiered one), the Quick Reference row, the Grok entry and the routing framework are corrected. Grok Imagine Video 1.5 Lite is added to the Grok entry from xAI’s own page ($0.02 / $0.03 / $0.14) and Artificial Analysis’s text-to-video board (#17 with audio, 993), with Runway’s note that its 1080p output is upscaled from 720p attributed to Runway; the Runway entry gains the October 1 Runway Dev listing (Weekly Roundup — October 12, 2026).
• What’s Coming — Grok Imagine 1.5 at 1080p marked resolved: the item still said both Grok models top out at 720p; native 1080p shipped July 31, 2026, now a priced tier.
• Artificial Analysis boards re-read, both toggles on all three pages — video-editing is now AA-Video-Editing v1.1 and the order changed: Wan 3.0 1,187, Dreamina Seedance 2.5 1,151, Omni Flash 1.1 1,139, H3 at 768p 1,119; on No Audio Omni Flash 1.1 leads (1,157) over Wan 3.0 (1,146). H3 is fourth, not second, and Omni Flash no longer shares first. Text-to-video v2.0 figures moved by one to four points (Utopai X 1,149, Seedance 2.5 1,143, H3 1,137, H3 Max 1,130; silent H3 Max now ahead of H3, 1,118 to 1,117). Image-to-video now has Vidu Q4 Preview third (1,179, provisional, in a rank 2–6 tie), and the no-audio note’s fifth entry is Vimoo 1.0 (1,348), not PixVerse V6. Banners: text-to-video 6 added (Grok Imagine Video 1.5 Lite, H3 Max, Utopai X and three more), image-to-video 1 (Vidu Q4 Preview), video-editing 2 (Seedance 2.5, H3 768p). About twenty secondary rank and Elo mentions are refreshed.
• Vidu Q4 Preview added (Quick Reference row and What’s Coming): image-to-video and reference-to-video only; 9 / 19 / 24 / 38 / 78 credits per second at 540p to 4K on Vidu’s own price table; the “from $0.014/sec” launch figure is carried as Vidu’s promotional claim and not as a price (Weekly Roundup — October 12, 2026).
• Kandinsky 6.0 Video added to Other Notable Open-Source Models: MIT License, 3B and 29B parameters, five-second clips with synchronized audio; the main Pro checkpoint on Hugging Face is behind a contact-information form, the distilled Pro and Lite repositories are not. The MAGI-2 “only license with no territory exclusion” line is reworded.
• Kling — 4.0 status refreshed; API prices for 3.0 Turbo, 3.0 and Omni added: the release note now carries the full 4.0 table and puts HDR output and two-minute extension under “coming soon”; the API page still lists no 4.0 or Flash price. The lines saying Kuaishou had published no Turbo or Omni price, and giving Turbo as 480p–720p, are corrected from the API price page as read today; the date those rows appeared is not known.
• Gemini Omni Flash — SynthID Detector added to the watermark line: public since October 7, images, audio and video, detects only SynthID marks from adopting companies, terms bar programmatic submission.
• Regulatory — SB 1000, AB 2713 and SB 1111 re-read against the chaptered texts; verification-tool duty added: every figure and date in the provenance item matched; the SB 1000 § 22757.2 duties (free public tool, upload or URL, API, third-party route) are added beside Google’s Detector, with no claim about whether the Detector is the tool the statute means.
• Regulatory — EU Digital Omnibus (Regulation (EU) 2026/1744) added; Article 50 item re-sourced to EUR-Lex: new Article 5(1)(ba) and (bb) prohibitions from December 2, 2026, the Article 50(2) grace to December 2, 2026 under Article 111(4), and high-risk dates of December 2, 2027 and August 2, 2028, all against the Official Journal text. Correction: the Code of Practice item said Google and Meta were the only AI-video-relevant signatories; the Commission’s text of July 31 names Black Forest Labs among Section 1 signatories. The Article 99 fine wording now says “whichever is higher”.
• MiniMax H3 — license caution reworded: the Community License withholds the grant in the US, EU, UK and South Korea (§ V.4) and invites separate requests from there (§ II); it also requires separate written authorization above $20 million in yearly revenue. “Not usable at all” is replaced.
• Claude Motion added to the Agentic Layer’s What to watch: Anthropic’s October 8 beta (Team and Enterprise) exports MP4 and says it uses no video generation model; Runway is named a launch partner on Runway’s own page.
• Checked, no change: Luma’s pricing page lists Seedance 2.0, 2.5 and Mini in credits per second; the Luma entry carries no third-party pricing and none is added. Runway’s changelog (newest October 2) and API changelog (newest October 1) re-opened; nothing dated October 3–11.

October 4, 2026

• Weekly Roundup R#24 paired refresh: lastmod → 2026-10-04 (lead-by-one before the October 5 Roundup); Mode B infographic generated. The “Last updated” line, which still read September 6, now reads October 4.
• Artificial Analysis rebuilt text-to-video on September 30 — every text-to-video Elo on this page is rescaled, and video-editing and image-to-video stay on v1.0: AA-Video-T2V v2.0 and T2V-Silent v2.0 judge every model at 1080p on a new prompt set with Kling 3.0 1080p (Pro) fixed at 1,000. With audio: Wan 3.0 1,156, Utopai X 1,148, Dreamina Seedance 2.5 1,142, MiniMax H3 at 768p 1,138, H3 Max 1,134. Silent: Wan 3.0 1,129, H3 1,118, H3 Max 1,115, Seedance 2.5 1,102, FLUX 3 1,094. Video-editing is still AA-Video-Editing v1.0 on both toggles (Wan 3.0 1,188; no audio, Omni Flash and Wan 3.0 level at 1,142), read from the page’s own payload for the first time on the No Audio toggle; image-to-video is still v1.0 with AA saying v2.0 is coming. Historical v1 figures are tagged in place, the leaderboard note is rewritten, and the Quick Verdict, Quick Reference and per-model entries carry the v2.0 numbers (Weekly Roundup — October 5, 2026).
• Gemini Omni Flash 1.1 and FLUX 3 are now on a board: the v2.0 text-to-video board scores Omni Flash 1.1 by name (#8 with audio, 1,113), so the entry’s line that AA had not scored it separately is replaced; FLUX 3 is #6 with audio (1,127) and #5 silent (1,094), which fires a second of the four Black Forest Labs reopen signals. The question is reopened, not re-answered.
• MiniMax H3 Max — the promotion changed a fourth time, and the Director minimum moved: fal’s H3 Max and Turbo pages, fetched October 4, show 40% off to October 15 ($0.048 / $0.03 / $0.096 and $0.024 / $0.015 / $0.048 per second at 768P / 480P / 1080p), replacing 50% off to September 30; list rates are unchanged. The Director page now quotes $0.08 per second with a $4.80 minimum per session (it read $1.20 on September 27) and no longer quotes $0.02. Also added: fal Recast (October 1, powered by H3 Max, $0.30 per second at 768p) and first-, middle- and last-frame control for Reference-to-Video (October 2) are added.
• MiniMax H3 — HeyGen Video, Boreal-H3 and Utopai X added as the eighth through tenth outside builds (all September 30): HeyGen says on its own catalog page that its model is built on H3 and post-trained by HeyGen; Creatify’s Boreal-H3 is post-trained for advertising on vendor-reported numbers; Utopai X debuted #2 with audio, with the H3 base attributed by Artificial Analysis, not by Utopai. MiniMax’s account quote-posted HeyGen and Creatify. The entry also links the open-weights claims tracker at the Viggle-Animate sentence (Weekly Roundup — October 5, 2026).
• MiniMax H3 Max — the “Base model” line corrected: it said H3 Max carried none of the Community License’s territory exclusions because fal serves it hosted; nothing supported that, and the license text bars use or display of the model or its Outputs in the excluded territories, derivatives included. fal is a San Francisco company and has not said publicly whether it holds a separate MiniMax license.
• Kling 4.0 added to What’s Coming — announced, not launched: Kling said on September 28 that the full model arrives this October, with Flash live for annual Ultra subscribers; the specs are vendor-only, no board lists it, and a Kling creative partner’s October 2 article calls the current access a closed beta. A pointer is added to the Kling entry.
• Runway — OpenAI Marketplace, Runway Ads, Project Continuum and a changelog sweep: the Marketplace listing (September 29; “launch partner” is Runway’s wording, not OpenAI’s directory’s), Runway Ads (September 30; a pilot, with results that are Runway’s own), Continuum (October 1; demos only, X-only), and the changelog entries for September 29 – October 2 (ElevenLabs V4, OpenAI Dots, Ideogram 4.5 in MCP, Seedance 2.5 Draft Mode in Agent). The Draft mode line no longer says it has no changelog entry (Weekly Roundup — October 5, 2026).
• Higgsfield — the cashback is now a standing tiered programme: the September 26 post said unused cashback expired September 30; Higgsfield’s API console, read October 4, describes only monthly cashback of 5% at $5,000 of spend rising to 20% at $50,000, and does not mention the 100% launch rebate.
• Regulatory — SB 1000 (Chapter 861), AB 2713 (Chapter 856) and SB 1111 (Chapter 862) signed September 30: three unsigned or pending lines are rewritten from the chaptered text; SB 1000 is an urgency statute in force now. Correction: Connecticut SB 5 is Public Act 26-15, signed by the Governor on May 27 per the General Assembly’s bill history; this page said it awaited signature. The October 1, 2026 provenance date has passed, and the October 1, 2027 detectability date this page carried is removed as unconfirmed.
• Currency fixes: the gemini-omni-flash-preview deprecation line (Google’s deprecations page no longer lists it; the Veo 3.1 preview IDs now shut down October 22); the HappyHorse days-since-launch counter, frozen at 69, now 161; the Wan 2.7 re-filing pointer, which still pointed at a tracker that published September 24; and Wan 3.0’s 30% discount, rechecked against Alibaba’s pricing page stamped October 3 and still live.
• Price superlative dropped: the “cheapest audio-capable API” claim for Grok Imagine is removed from the intro, the Quick Verdict, the Quick Reference, the Grok entry and the routing framework, which also printed $4.20 a minute where the page elsewhere gives $4.80. Grok Imagine Video 1.5 is now stated plainly at $0.08 per second with native audio, and the routing framework’s “cheapest API” question is answered by tier from this page’s own prices.
• ComfyUI, fal Agent and LTX added: Comfy Router (September 23) and Comfy API (September 30) in the Agentic Layer’s What to watch; fal Agent as a What-to-watch line only, with the six-agent table unchanged; LTX for Post-Production (September 29 – October 2) in the LTX entry, from the vendor’s release notes.
• Changelog trim: September 10, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: October 4 / September 27 / September 20 / September 13.

September 27, 2026

• Weekly Roundup R#23 paired refresh: lastmod → 2026-09-27 (lead-by-one before the September 28 Roundup); Mode B infographic generated.
• OpenAI — the Sora API removal executed: the deprecations page, fetched September 27, still lists September 24, 2026 as the shutdown date for the Videos API, sora-2, sora-2-pro and three dated snapshots, with no replacement in any row. The Sora entry’s status line and the OpenAI What’s Coming item move to the past tense (Weekly Roundup — September 28, 2026).
• Meta — Muse Realtime Avatar added to the Muse What’s Coming item (September 23): 448×768 at 25fps, about 870ms to first byte, 120 → 2 function evaluations per chunk, 12 sessions per GB200, Video Seal watermark, all per Meta’s post. Meta’s rater study against Runway Characters and HeyGen LiveAvatar is recorded as Meta’s, with its one statistical tie and no outside replication. Muse Video’s Design Arena position refreshed to #9 (1,456, board dated September 21).
• HeyGen HyperFrames — Code2Video added (report dated September 18, launched on X September 21): 168 briefs, five axes, a purpose-trained judge at 82% human agreement against 75% for general vision-language models; top four within 12 Elo, open-weight models within 50. Scoped as an assembly-layer benchmark, not a generation one (Weekly Roundup — September 28, 2026).
• Runway — five changelog entries and one X-only launch: Grok Bot plugin (September 17), four Kling models on Runway MCP (September 18), Brand Kits readable via MCP (September 22), DaVinci Resolve plugin and Cursor Marketplace listing (September 23), all from the vendor’s dated changelog; Draft mode for Seedance 2.5 (September 24) from @runwayml, with no changelog entry. The September 17 and 18 entries predate the September 20 refresh and were missed by it.
• Seedance 2.5 — a first-party Draft mode, and the 1080p discount has lapsed: ByteDance’s BytePlus page documents a two-step Draft mode (draft billed as 480p, final at target resolution); Pika, Higgsfield and Runway each launched it within about 41 hours, September 23–24. The same page, fetched September 27, no longer lists the dated 28% 1080p discount that ran to September 17; the effective-rate note is marked Superseded in place.
• Higgsfield — a self-reported $1B run rate, Preview Mode and an API cashback offer: the September 24 claim that annualized revenue crossed $1 billion is recorded in the Supercomputer entry as the company’s own, unaudited, beside the documented $700M at the August 17 close; Seedance 2.5 Preview Mode (480p draft, 1080p render, September 23); 100% instant cashback on API spend, up to $100,000 per business from a $20 million pool (September 24), with later posts raising the ceiling to $200,000 and describing the rebate as API credits. No terms page states expiry or eligibility.
• Pika — Seedance 2.5 Draft Mode (September 23) and the API offered to Grok Bots (September 24), both from @pika_labs.
• MiniMax H3 — HyperFlow added as the seventh outside build: Video Rebirth’s 8-step LoRA for H3 (September 24); the Hugging Face card states the Community License’s territorial scope, the announcement post does not. Cross-referenced from the Bach 1.0 Pro entry.
• MiniMax H3 Max — two more fal variants; the promotion held: 3D to Video (September 23) and Extend Video (September 24), per @fal; H3 Max and Turbo still 50% off to September 30 on fal’s model pages, fetched September 27. The Director page still quotes $0.02/sec beside a promotion it says ended September 14; the entry says so and points to the $0.08 list rate instead of carrying a rate dated September 13.
• Black Forest Labs — FLUX 3 Action noted; Design Arena refreshed: @bfl_ai’s open-weights 7B World Action Model for robot control (September 23) is recorded as robotics, outside the reopen test; FLUX 3 Video sits #3 (1,493, Preliminary) on the Design Arena board dated September 21, up from #4.
• Regulatory — SB 1000 still unsigned on September 27, per the leginfo record: last action remains “Enrolled and presented to the Governor” on September 2, with the September 30 deadline under Article IV, Section 10(b)(2) standing.
• Leaderboard refresh to September 27, 2026 — both toggles read on all three scored pages; no new top-eight entrant: With Audio — text-to-video unchanged (Omni Flash 1,233, Wan 3.0 1,229, H3 Max 1,227, H3 1,220, Seedance 2.0 720p 1,210); image-to-video H3 Max 1,194, H3 1,181, Omni Flash 1,178, Seedance 2.0 720p up to #4 (1,176), HiDream-O1-Video down to #5 (1,174); video-editing Wan 3.0 1,188, H3 1,132, Omni Flash 1,125, HappyHorse-1.0 1,089, Wan 2.7 1,077. No Audio — text-to-video Wan 3.0 1,335, Omni Flash 1,332, H3 1,302, HappyHorse-1.0 1,286, HappyHorse-1.1 1,272; image-to-video Omni Flash 1,369, Bach 1.0 Pro and Wan 3.0 1,362, H3 1,358, PixVerse V6 1,340; video-editing, recorded here for the first time, Omni Flash and Wan 3.0 tied at 1,142, Wan 2.7 1,116, H3 1,100, HappyHorse-1.0 1,089. “Added in the last month” banners: HiDream-O1-Video (image-to-video, With Audio) and Bach 1.0 Pro (image-to-video, No Audio); none on text-to-video or video-editing. Bach 1.0 Pro now carries a board release date (September 2026) and price ($7.20/min).
• Quick Reference anchor fixed: the {#quick-reference} id sat on a bold paragraph, where it never became an anchor, so the H3 Max entry’s in-page link to the leaderboard note went nowhere; the id now sits on the Quick Reference heading.
• Stack exit test (fourth against this page) — one staleness fix, from RCTV’s own fetch; nothing adopted on the comparison’s word. (1) Andersen v. Stability AI: this page said the court’s recent-filings list ran through August 31; the court’s own page, fetched today, runs to September 25 — the plaintiffs’ September 21 motion to modify the schedule (relief redacted), Stability’s September 25 opposition, a sealed September 22 sanctions reply against DeviantArt. The training-data item is updated from that read. (2) Not adopted, unverified: an October 20 sanctions hearing and an April 5, 2027 trial date (neither on the court’s page); a Reuters account that the Disney–Sora deal never funded; MLex/GCR reporting that NVIDIA is weighing a voluntary EU Article 22 referral for Hugging Face (paywalled) — all queued for Writer-Researcher verification. (3) Board figures matched RCTV’s own September 27 read on both boards it reports. (4) Comparison wrong on inspection: it lists Pika, Runway, Meta, Black Forest Labs and ByteDance as verified quiet in the week of Seedance 2.5’s Draft mode, Runway’s five changelog entries, Muse Realtime Avatar and FLUX 3 Action.
• Changelog trim: September 6, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: September 27 / September 20 / September 13 / September 10.

September 20, 2026

• Weekly Roundup R#22 paired refresh: lastmod → 2026-09-20 (lead-by-one before the September 21 Roundup); Mode B infographic generated. Ships Sunday per the Stack-paired-with-Roundup cadence.
• OpenAI — the Sora API removal is confirmed on the vendor’s own deprecations page: sora-2, sora-2-pro and every dated snapshot are removed September 24, 2026, developers were notified March 24, and the recommended-replacement column is empty for every Sora row. The Sora entry and the OpenAI What’s Coming item now say so; the second and last stage of the shutdown (Weekly Roundup — September 21, 2026).
• Runway — six dated entries added: Runway Dev MCP (September 2, a hosted MCP server for coding agents; a pulse miss recorded here on the vendor’s own dated index); the model-licensing post (September 11) with the licensing page linked — the page carries no date, so fresh-offer-versus-repromotion of the August 20 enterprise post stays unresolved and the line is dated to the post; the “Fall 2026 collection” (September 16, five models, none Runway’s: MiniMax H3 Max, Flux Video Upscale and Edit, Fish Audio S2.1 Pro, Cartesia Sonic 3.6); Moderation in Real Time (September 16, Zentropi’s CoPE-B adapter over Google’s Gemma-4-26B-A4B-it, an average review under 0.5 seconds, framed as readiness for real-time models still to come); Enhance Frame Rate (September 17, 25–120 fps plus NTSC 59.94, one credit per two seconds); unified pricing (September 18, one credit pool across the app and Runway Dev, 10% enterprise signing bonus to December 31). The licensing post is an R#21-window item routed here under the out-of-window rule; the brief’s adversarial verifier had fetched the wrong September 11 post and called it misattributed — it was not.
• Higgsfield — API launched (September 16 announced, September 17 live): “50+ frontier models in one API” is the pitch; the console’s own catalog listed 22 models on September 20, 14 of them video, with per-model discounts (Kling 3.0 50%, Seedance 2.5 30%, MiniMax H3 45%), and docs.higgsfield.ai still documents only the Soul family. Recorded in the Supercomputer entry as the claim beside the visible count (Weekly Roundup — September 21, 2026).
• Pika — relaunched September 17 and repriced: tiers now Free / Starter $10 / Creator $35 / Fancy $95 / Enterprise with “All models” on every paid tier at 60 credits per dollar (the entry read “From $8/mo” until today, superseded in place); the pika.art menu lists Seedance 2.5, GPT Image 2.5, Grok Imagine 1.5, Gemini 1.1 Omni Flash, Nano Banana 2 and MiniMax H3 beside Pika 2.5. Automatic model routing, Swap Anything and Camera Director are third-party claims with no Pika primary and are not carried. Quick Reference row updated.
• Veo — “Love, Rendered” (September 9) added: Google’s third public Performance Control showcase, a Primordial Soup documentary reconstructing a couple’s youth from restored photographs; an R#21-window item routed here under the out-of-window rule.
• Regulatory — California SB 1050 signed September 16, Chapter 246, Statutes of 2026: synthetic-performer disclosure in advertisements, verified against the leginfo record and the chaptered text — a synthetic performer is by definition not an identifiable natural person, disclosure wording is prescribed, expressive-work ads are exempt, no urgency clause so operative January 1, 2027. SB 1000 remains unsigned as of September 20 with the September 30 constitutional deadline standing; the provenance-mandates item is rewritten accordingly.
• MiniMax H3 Max — the promotion changed a third time: fal’s model pages, fetched September 20, now show 50% off to September 30 on H3 Max ($0.04 / $0.025 / $0.08 per second at 768P / 480P / 1080p) and Turbo ($0.02 / $0.0125 / $0.04), replacing the 75%-off rates that ran to September 14. The earlier figures were right when written and are preserved in place; the Director page was not re-fetched and its rate is marked as dated September 13. H3 Max also joined Runway’s Fall 2026 collection on September 16.
• Leaderboard refresh to September 20, 2026 — no new top-eight entrant, and the text-to-video lead changed hands inside the error bars: Gemini Omni Flash #1 T2V (1,233) over Wan 3.0 (1,229), reversing September 13’s three-point Wan lead — the two have swapped twice in a week and the Quick Verdict now calls them level rather than ranking a tie; Wan 3.0 holds video-editing (1,190) and sits #6 I2V (1,164); H3 Max #1 I2V (1,195), #3 T2V (1,227); H3 #4 / #2 / #2 (1,220 / 1,181 / 1,132); Seedance 2.0 720p #5 / #5 / #6 (1,210 / 1,174 / 1,035); Omni Flash #1 / #3 / #3 (1,233 / 1,177 / 1,126); HiDream-O1-Video still #4 I2V (1,176); Grok Imagine Video 1.5 #8 I2V (1,099); MAGI-2 Preview #6 T2V (1,156, up from its #9 debut) and #9 I2V (1,096). All from RCTV’s own fetch of the three boards.
• Fourth Cowork exit test against this page — one genuine gap found and confirmed to primary; nothing adopted on the comparison’s word. (1) Video Rebirth’s Bach 1.0 Pro sits #3 on Artificial Analysis’s no-audio image-to-video ranking (Elo 1,359) — a vendor this page had never tracked, on a board it reads but had only been reading on the with-audio toggle. Confirmed against AA’s own page and the vendor’s May 7 PR Newswire release, added above as a What’s Coming watch entrant; the Quick Reference’s no-audio note was stale (it said Omni Flash led no-audio text-to-video at 1,324; Wan 3.0 leads at 1,336) and is corrected. (2) Comparison wrong on inspection, three times: it places Video Rebirth in Hong Kong (the company’s own release is datelined Singapore and says so); it records H3 Max’s promotion as lapsed (fal’s page, fetched today, says 50% off to September 30); it logs fal’s Reference-to-Video as “four weeks overdue” (generally available since September 10 per fal’s own post, carried here since September 13). It also lists Runway and Pika as “verified quiet” in a week Runway published four dated posts and Pika relaunched. (3) Not adopted, unverified: the “(0912)” MAGI-2 entry carrying no Open Weights badge; a §17500 misdemeanour reading of SB 1050; the sample-pool revisions on Omni Flash; an April 5, 2027 Andersen trial date (this page deliberately carries none). (4) Queued for Writer-Researcher verification: the MAGI-2 badge claim; SB 1050’s penalty provision; Bach 1.0 Pro’s specs and price. (5) Their with-audio board figures matched RCTV’s own September 20 fetch to the point.
• Black Forest Labs — Runway widened its shelf, not a fifth aggregator: Runway’s September 16 collection added Flux Video Upscale and Flux Video Edit; the reopen test’s “fifth aggregator” trigger has not fired, and the entry says so.
• Changelog trim: September 1, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: September 20 / September 13 / September 10 / September 6.