AI Video Stack 2026 — Changelog Archive

This is the complete edit history for AI Video Stack 2026: Live Comparison & Rankings. Entries run newest-first and are preserved exactly as written — this is an audit trail, not a summary. The four most recent updates also appear on the main Stack page.

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.

September 13, 2026

Weekly Roundup R#21 paired refresh: lastmod → 2026-09-13 (lead-by-one before the September 14 Roundup); Mode B infographic generated. Ships Sunday per the Stack-paired-with-Roundup cadence.
CORRECTION — Wan 2.7 is not open-weights, and this page said it was: the entry carried License: Open source from April 3, 2026 with no vendor claim behind it. 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 entry now reads API/cloud only; the Quick Reference row and the cross-references in the Wan 3.0 and NVIDIA–Hugging Face items are corrected with it, and the open-weights line is recorded as stopping at Wan 2.2. Surfaced by the Writer-Researcher’s open-weights audit of September 13 and confirmed against the three hosts directly; re-filing the entry out of the open-source section is deferred to the open-weights tracker piece due September 24.
Adobe and Runway in the timeline (September 8): Adobe’s native Generative Media Tool in Premiere Pro and After Effects names Firefly plus Veo, Kling, Runway and Luma as partner models; Runway’s own free plugin into the same two apps carries Gen-4.5, Seedance 2.5, Kling 3.0 Pro and Veo 3.1, billed on the user’s Runway credits. Recorded in the Adobe Firefly, Runway, Veo, Kling, Luma and Seedance entries and the Quick Reference; the roster discrepancy is carried as an observation, not explained (Weekly Roundup — September 14, 2026).
Kling — state fund and CP Robot take stakes (August 31 filing): Kuaishou’s stock-exchange announcement, as reported by TechNode on September 1, brings the China Artificial Intelligence Industry Investment Fund in at RMB1.4 billion for about 1.14% and Charoen Pokphand’s CP Robot at about $19.29 million for about 0.11%, both with redemption rights. Carried from the reporting with the filing linked; the filing PDF did not render for direct reading this pass and is queued for Writer-Researcher verification. An R#20-window item routed here under the out-of-window rule.
MiniMax H3 — Viggle-Animate added as the sixth outside build (September 4): Viggle’s “first open-weight model” replaces a character in a video from a single repainted frame, 124 frames in 26 seconds on one B200 per the model card. The card states the weights are a Model Derivative of MiniMax H3 under the MiniMax H3 Community License — the territory clause that excludes the US, EU, UK and South Korea — which Viggle’s announcement does not mention. An R#20-window item routed here under the out-of-window rule.
MiniMax H3 Max — 1080p tier, promotion extended, three more fal ships: fal’s model pages now list a 1080p tier on H3 Max ($0.16/sec list) and Turbo ($0.08/sec list), and the 75%-off launch rates on both now end September 14, not September 7 as this page said on September 6 — fal moved the date; the earlier figure was right when written. Director 1.1 (September 8: keyframe insertion mid-generation, audio at any point, native 1080p at twice the rate, sessions up to 15 minutes by default, $1.20 minimum), Multi Angle (September 9) and Reference-to-Video generally available (by September 10) added to the entry. The 768P-only resolution note is superseded in place.
Leaderboard refresh to September 13, 2026 — one new top-five entrant: HiDream-O1-Video (HiDream.ai, Beijing) debuts #4 image-to-video at Elo 1,186 with no first-party launch announcement found; added to What’s Coming as a watch entrant. Wan 3.0 now leads text-to-video by three points over Gemini Omni Flash (1,240 vs 1,237, inside the error bars) and holds video-editing (1,197); H3 Max #1 I2V (1,206), #3 T2V (1,230); H3 #4 / #3 / #2 (1,225 / 1,190 / 1,129); Seedance 2.0 720p #5 / #2 / #6 (1,220 / 1,197 / 1,039); Omni Flash #2 / #5 / #3 (1,237 / 1,181 / 1,122); Grok Imagine Video 1.5 #7 I2V (1,115); MAGI-2 #9 I2V (1,100). Everything except the HiDream entry moved within noise.
Routine freshness across five entries: Black Forest Labs — FLUX Video Edit (September 10), $0.03/sec of output, about 50 seconds for a 10-second clip, eight edit types, live on fal the same day; @bfl_ai posting again is one of the four reopen signals the September 10 analysis named, so the Big Ten question is reopened, not re-answered. HeyGen — an open-source real-time demo stack (September 10) pairing LiveAvatar and HyperFrames with OpenAI’s GPT-Live-1 voice model. Seedance 2.0 — available to US companies on fal, hosted on US infrastructure (September 10). Grok Imagine — first-and-last-frame control and seamless loops (September 8). Regulatory — SB 1050 presented to the Governor September 9; both bills unsigned as of September 13; under Article IV, Section 10(b)(2) of the California Constitution each becomes law on September 30 if neither signed nor vetoed.
Third Cowork exit test against this page — nothing adopted on the comparison’s word; two board positions refreshed from RCTV’s own fetch. (1) The comparison’s HiDream-O1-Video, Wan 3.0 and H3 Max positions matched RCTV’s September 13 fetch and were recorded from ours. (2) Two RCTV lines were stale and are refreshed from our own board read: LTX-2.5 (24th and 26th of 36, Fast above Pro) and Agnes-Video, where Sapiens’ 2.5 tier now sits #19 at $1.50/min. (3) Comparison wrong on inspection, twice: it still labels Wan 2.7 “Open source” with “downloadable checkpoints in general circulation” — the correction above stands on three hosts checked directly; and it says H3 Max’s promotion “was due to lapse around September 7” at 50% off — fal’s page, fetched today, says 75% off through September 14. (4) Not adopted, unverified: HiDream-O1-Video at $5.80/min (the model page 404s); an Artificial Analysis sample-count anomaly the comparison reports across three models. (5) Queued for Writer-Researcher verification, not carried: a Minnesota statute holding AI platform owners liable for nonconsensual intimate imagery; Seedance 2.0 Pro “native 4K/10-bit (June 23)”; the Kuaishou filing PDF. (6) Still ahead: the comparison spent its lead story on an Andersen v. Stability trial date that RCTV’s September 7 docket check had already shown does not exist, and recorded Kling and Black Forest Labs as “verified quiet” in a week that carried the Kling filing and FLUX Video Edit.
Same-day amendment from the open-weights audit — two “Other Notable Open-Source” bullets corrected: SkyReels V4 is preview-only with no published weights — no V4 repository in Skywork’s GitHub org (a February 26 request for the weights sits unanswered), no Hugging Face listing, and the skyreels.dev free tier offline; this page said open-source weights were available for local deployment, and they never were. HY-World 2.0’s licence terms recorded from the licence text: EU, UK and South Korea excluded, separate licence above 1M monthly active users. Both surfaced by the Writer-Researcher’s audit for the open-weights claims tracker (publishes September 24) and confirmed directly.
Changelog trim: August 18, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: September 13 / September 10 / September 6 / September 1.

September 10, 2026

Black Forest Labs — Big Ten question closed for now, watch-entrant status held: the September 10 analysis answered the question the What’s Coming entry had carried open since August 18. The entry now carries the verdict and its link in place of the “scheduled analysis” language, plus the four reopen triggers the piece names (a fifth aggregator, an enterprise or funding disclosure, an Artificial Analysis board entry, @bfl_ai posting again). Standalone pass, lastmod → 2026-09-10; no board, spec-card or roster changes — the paired refresh for Roundup #21 follows on September 13.
Changelog trim: August 16, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: September 10 / September 6 / September 1 / August 18.

September 6, 2026

Weekly Roundup R#20 paired refresh: lastmod → 2026-09-06 (lead-by-one before the September 7 Roundup); Mode B infographic generated. Ships Sunday per the Stack-paired-with-Roundup cadence.
MiniMax H3 Max — two new fal variants on the same weights: H3 Max Turbo (September 2) — fal’s “2× the speed at half the cost” preview, list $0.04/sec at 768P and $0.025/sec at 480P per fal’s own model page, absent from every Artificial Analysis board; carried as an unranked preview, and its “97th percentile” quality target attributed to fal rather than adopted. H3 Max Director (September 3) — the continuous, action-controlled model behind fal.live, now a paid API at $0.02/sec promotional through September 14 ($0.08/sec list, 60-second minimum). The 75%-off launch rates on H3 Max and Turbo both end September 7.
fal.live status recorded, not narrated: two of seven channels running on September 3, six of nine on September 6 (two channels added, three still paused), no statement from fal on what failed or what fixed it. The H3 Max entry now carries the recovery alongside the pause.
MiniMax H3 — H3-World added to the entry: a research team unaffiliated with MiniMax or fal (Xingyi Yang, PolyU/NUS/Oxford, September 2) converted the open H3 into a world model on 8,000 training samples. Carried as what it is — no product, no independent benchmark — and as the fifth outside build on one open release inside five weeks.
Runway — Solaris and GWM Worlds 2 added: two request-form research previews from one world-model lineage. Solaris (August 31) generates interfaces, not video, and the entry says so first; GWM Worlds 2 (September 3) generates continuous 720p/24fps video with 48kHz audio and is the successor to the GWM-1 line behind Characters. Runway’s 61–71% preference figures for Solaris are attributed to Runway’s own study, not to the secondary that misattributed them. The Cosmos Coalition What’s Coming item now points at both.
Regulatory — SB 1000 presented to the Governor September 2; SB 1050 Enrolled September 4: both verified against the leginfo primary record on September 6, both unsigned, neither law. SB 1050 (synthetic-performer disclosure in advertisements, Assembly concurrence 40–0 on August 31) joins the provenance-mandates item as California’s counterpart to New York’s synthetic-performer law.
Leaderboard refresh to September 6, 2026 — no new top-eight entrant; two positions added: Wan 3.0’s first image-to-video score (#5, Elo 1,175) and MAGI-2 Preview’s first text-to-video score (#9, 1,114; I2V now #8, 1,100) recorded in their entries, both from RCTV’s own board fetch. Wan 3.0 and Gemini Omni Flash are level at Elo 1,238 on text-to-video and the page now says “level” rather than ranking a tie; Wan 3.0 holds video-editing outright (1,190). MiniMax H3 Max #1 image-to-video (1,201); MiniMax H3 #3 I2V (1,187) and #2 video-editing (1,130); Seedance 2.0 720p #5/#2/#6 (1,222 / 1,191 / 1,040). Every Elo moved by one to three points — noise, updated for currency only.
Runway Dev API — two dated entries from the API changelog, the surface added to the registry on August 30: ACEScg OpenEXR delivery for HDR outputs (August 31), a compositing-pipeline format carried in the Runway entry as tooling rather than capability; and MiniMax H3 Max available on Runway Dev (September 3), added to the H3 Max entry — the first time a second vendor distributes a third party’s post-train of another lab’s open weights. The H3 Max item was not in the R#20 brief or the Saturday draft; the changelog was read at Thursday’s sweep and not again until Sunday integration.
Second Cowork exit test against this page — one genuine miss, one conflict resolved in RCTV’s favour, nothing adopted on the comparison’s word. (1) World Labs Atlas (September 1) was absent from RCTV’s pulse, brief and draft — no monitoring lane covers World Labs. Verified against World Labs’ own post and added above as a What’s Coming item and to the September 7 Roundup; a discovery gap, scored as such in the autonomy ledger. (2) H3 Max promotional rate: the comparison says 50% off at $0.04/sec; fal’s model page, fetched September 6, says 75% off at $0.02/sec ending September 7 — this page’s September 1 correction stands. (3) The comparison’s Wan 3.0 image-to-video and MAGI-2 text-to-video positions matched RCTV’s own fetch and were recorded from ours, not theirs. (4) Queued for WR verification, not carried: Stability AI’s reported $76M round of August 25 (out of window, relevant to the training-data item); H3 Max’s signed-in free tier (the comparison says five 15-second clips, this page says fifteen 10-second clips). (5) Still ahead: the comparison still labels LTX-2.5 as Lightricks under Apache 2.0, and still publishes an Andersen trial date this page deliberately withholds.
Black Forest Labs — freshness ahead of the September 10 analysis: Pika’s integration date corrected to August 4 (same day as Runway, per Pika’s own post timestamp), fal.ai recorded as the fourth aggregator (serving FLUX 3 Video by August 26, per fal’s model page and its own dated pricing card), Luma’s FLUX Upscale pickup (September 1) added, and the Design Arena figure moved from the #2 debut to the live #4 (Elo 1,494, “Preliminary”). The Big Ten question stays open on the page until the analysis publishes.
Changelog trim: August 11, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: September 6 / September 1 / August 18 / August 16.

September 1, 2026

Catch-up refresh — this page missed the R#18 and R#19 paired updates. The standing cadence is that the Stack refreshes alongside every published roundup. It did not happen on August 23 or August 30, and nothing caught it: the changelog invariant check asserts structure (four inline blocks, each mirrored in the archive) and holds perfectly on a page nobody touched. Two roundups shipped against a lastmod of August 18. What that cost readers: this page claimed Gemini Omni Flash had swept all four Artificial Analysis boards and that MiniMax H3 led video-editing. Both were true in July and neither was true by August 19. A freshness check now runs in rctv-gates and in the daily task sweep, comparing lastmod against the newest published roundup.
Wan 3.0 — Alibaba (NEW ENTRY, commercial) and the section is now the Big Ten: Alibaba opened an invite-gated commercial beta August 6, 2026; it first charted August 19 and went straight to #1 on Artificial Analysis text-to-video (Elo 1,241) and #1 video-editing (Elo 1,189), displacing Gemini Omni Flash and MiniMax H3 respectively, and has held both. 30-second single-pass generation, document-to-video input (doc/xls/ppt/pdf/md up to 100MB), 480p/720p/1080p, no 4K. Closed — no published weights, no GitHub listing, distinct from the open-weights Wan 2.7 line, which now carries a disambiguation note. Promoted on the criterion this page already states for watch entrants (“not a counted row until a category-leading model ships”); the section header moves from Big Nine to Big Ten and eight live references follow it. Historical changelog entries keep “Big Nine” as written, per the audit-trail contract. Distribution followed the established pattern: eleven surfaces in eight days (Runway Aug 24, Pika and ComfyUI within two days, then ArtArch, Media.io, OiiOii AI, Picsart, PowerDirector, A2E, PixelDojo, DeepInfra).
MiniMax H3 Max — post-trained by fal (NEW ENTRY, commercial): the first third-party fine-tune to top a leaderboard this page tracks — #1 image-to-video (Elo 1,202), #3 text-to-video (1,235), ahead of the official MiniMax H3 endpoint at #4. A 5-second 720p clip in under 3 seconds: generation faster than playback. The three speed multipliers are three different comparisons and are labelled as such (35× vs the official H3 endpoint per fal; 50× per Design Arena independently; 15× against a quality-matched group) — never blended. No per-second rate is published on fal’s own pricing surface, so none is carried. fal’s board label reads “#1 image-to-video with audio”; Artificial Analysis retired its with-audio boards on August 11, so the live #1 is on the plain board — noted in the entry. fal.live (August 31) and its “H3 Max Director” variant are recorded including the pause roughly three hours after launch.
Gemini Omni Flash — the July sweep is over, and Omni 1.1 Flash added: now #2 T2V (1,237), #4 I2V (1,179), #3 video-editing (1,124) as of August 28. Displaced rather than degraded — the Elo barely moved; two newer models scored higher. Omni 1.1 Flash (August 27) adds scene extension, first/last-frame specification, up to three video references, 4K upscale, and a 360p fast mode. The distribution is split and the entry says so: Google frames the full set as “available via APIs” for developers, while the Gemini app account scopes the consumer rollout to scene extension only. Artificial Analysis has not scored Omni 1.1 separately — the board entry carries none of AA’s version markers — so the figures above describe Omni Flash, not the update.
Higgsfield — the round closed, above the reported range: $400M Series B at $5.4B on August 17, DST Global leading with Goldman Sachs Alternatives, Valor Capital and Tribe Capital. Disclosed: $700M annualized revenue, 30M users across 200 countries, 390 Fortune 500 clients. This page carried the July report of “$300–500M at a $5B valuation” on a ~$500M run rate; the closed terms came in above the reported range on both, and the entry now says so rather than silently replacing the earlier figure.
Regulatory — California SB 1000 is Enrolled, not law: the Assembly concurred in Senate amendments 39–0 on August 27 and the bill was Enrolled August 30, 2026, now on Governor Newsom’s desk with an urgency clause, meaning immediate effect if signed. Verified against the leginfo primary record. What it would change to SB 942 is now stated (deletes the user-count threshold, renames the detection tool a “disclosure verification tool,” drops the manifest-disclosure option requirement, requires the latent disclosure to distinguish AI-generated from AI-modified). Still unsigned; the operative statute remains SB 942. Given the ten-week SB 1000 misattribution corrected on August 11, the status wording here is deliberately literal.
Regulatory — MPA and ByteDance sign (NEW What’s Coming item): the first AI-safeguards accord between Hollywood and a Chinese tech company, August 17, covering Seedance and Seedream across TikTok, the USDS joint venture, CapCut and Dreamina. Carried with the two qualifications that matter: it is a voluntary truce, not a licensing deal — studios receive no payment — and neither party published enforcement thresholds or detection mechanisms.
Routine freshness across five entries: Runway Ruby (August 21) — SDR-to-16-bit-HDR conversion up to 30s, ProRes/EXR, Max and Enterprise only, a colour-pipeline tool that moves no benchmark. HeyGen LiveAvatar (August 25) — concurrency cap removed entirely, “1 avatar or 10,000,” down to $0.01/min at scale. Seedance 2.5 — US rollout completed August 21, closing the regional sequence begun August 14; five platforms on 1080p within four days. Seedance 2.0 Pro — benchmark note rewritten to the live boards (#5 T2V, #2 I2V, #6 video-editing); the mid-July with-audio figures it carried no longer have a live board behind them. Black Forest Labs — FLUX Upscale (August 20), the first capability update since the July 23 launch, folded into the What’s Coming item as evidence for the scheduled analysis rather than as a new story.
Agentic layer — two additions to What to watch: Comfy MCP shipped local and open-source August 18, giving the self-hosted layer the same agent entry point the hosted platforms have — the first time this category extends below the API line. Tencent Hunyuan previewed HyCreator (August 20–21), claiming 10-minute films with “zero human intervention” plus a real-time interactive-editing mode, application-gated; carried as a claim, not a capability — no independent hands-on exists.
First Cowork exit test ever run against this page — three live errors found, all confirmed against RCTV’s own records rather than adopted on the comparison’s word. (1) MiniMax H3 Max resolution: this refresh first wrote 720p; R#19 had already verified 768p (1344×768) against fal’s model page seven times over, with fal’s own X post carrying the conflicting 720p. Corrected to 768p across four surfaces. (2) LTX-2.5 licence: the page called it Apache 2.0 in three places while the entry’s own licence bullet described a $10M-revenue threshold — a condition Apache 2.0 does not contain. Both cannot be true; the Apache label is removed, the revenue threshold retained, and the precise licence name marked pending primary verification. Same failure class as the MiniMax territory-exclusion catch: the release event was tracked, the licence text was not read. (3) Seedance 2.5 pricing: “no published consumer or developer price from any aggregator” is no longer safe — a first-party BytePlus rate card is reported to exist. Flagged, not adopted, pending verification against ByteDance’s own surface.
Wan 3.0 — the vendor-docs label conflict logged: trade coverage August 24–25 described an official release; Alibaba’s own Model Studio API reference still read “Currently in preview.” Tracked as preview until the vendor docs change. This is RCTV’s own R#19 discovery gap — no monitoring lane covers vendor API documentation — recorded on the page rather than left in the ledger.
Queued for primary verification, deliberately not carried: Gemini Omni 1.1’s chained-extension ceiling and whether its 1080p/4K are upscaled rather than natively rendered; the gemini-omni-flash-preview deprecation date; Wan 3.0’s per-second rate card and API-documented specs; H3 Max’s per-second rate; LTX-2.5’s licence text and arena scores against its 67% launch claim; Seedance 2.5’s first-party rates; FLUX Upscale pricing. Each would improve this page and none is verified to primary yet.
WR verification pass returned the same day — seven page-facing facts resolved to primary, three of them corrections to what this page carried hours earlier. Nothing below was adopted from the comparison document; every figure is from the vendor’s own surface, fetched 2026-09-01.
- Gemini Omni’s 1080p and 4K are UPSCALED, not natively rendered. Google’s own resolution table uses that word: “1080p output (upscaled),” “4K output (upscaled).” 720p is the only resolution Omni renders natively, which is the distinction that matters against Veo 3.1 and Kling 3.0. Also pinned: GA on 2026-08-27 as gemini-omni-1.1-flash, gemini-omni-flash-preview deprecated 2026-09-30, scene extension in 10-second increments to a 40-second cumulative ceiling (four chained requests, not one pass), and $17.50/1M video output tokens ≈ $0.10/sec.
- MiniMax H3 Max pricing is publishable after all, and our discount figure was wrong. fal’s model page rendered cleanly today when it would not for R#19 three days ago: $0.08/sec at 768P list, $0.05/sec at 480P, with a 75% promotional discount ending September 7 — not the “half price for 14 days” this page and the roundup both carried. R#19 declined to print an unverified per-second figure that has now proved correct; that is the discipline working, not a miss.
- CORRECTION to this page’s own entry, made hours after writing it: the “1344×768” pixel figure was unsourced. No surface states it; fal’s API schema offers 480P and 768P as labels only. It was an inference that read like a specification. Removed here and in the entry.
- LTX-2.5’s licence, verified against the licence text itself (LTX-2.x Community License Agreement, August 11, 2026): $10M revenue threshold aggregated across affiliates; derivatives defined to include fine-tunes, distillation and training on LTX-2.x-generated synthetic data; a competing-model training ban on commercial use; military/weapons/nuclear ban; two separate clauses barring removal of watermarking or provenance; and a stated intent to qualify under EU AI Act Article 53(2), carved out from 53(1)(c) and (d). Added alongside the measured board position — both tiers at Elo 1,060, ranks 21–23 of 33 — against the vendor-commissioned 67% win-rate launch claim.
- Seedance 2.5 first-party pricing exists and this page was wrong to say otherwise. BytePlus ModelArk publishes $0.103/$0.231/$0.569 per second for 480p/720p/1080p, with a dated 28% discount on 1080p only, August 14 through September 17, putting the currently-billed rate near $0.41/sec.
- Wan 3.0 list pricing confirmed at $0.05/$0.10/$0.20 per second with a live 30% discount the page now names, six endpoint regions rather than five, and the full API-documented spec set. Its “Currently in preview” label was re-checked against a page Alibaba itself stamped 2026-09-01 and still reads preview — trade coverage describing an official release remains uncorroborated by the vendor.
Leaderboard structure clarified, and a standing ambiguity closed. Artificial Analysis runs three scored leaderboard pages, each carrying a With Audio / No Audio toggle that swaps the whole ranking on the same URL. RCTV read that as “three boards”; the comparison document read it as separate audio and no-audio boards. Both were half-right, and the disagreement was really about whether different Elo per audio state implies a different board. It does not. The Quick Reference now states the structure and says which set this page quotes.
Training-data litigation added to What’s Coming — with no trial date, on purpose. Carlos ruled generative-media fair-use cases on-beat for the regulatory lane at the 2026-09-01 meeting, reversing a pulse exclusion. Andersen v. Stability AI is now carried. The circulating September trial date is not published: the court’s own case page has no trial-date field, govinfo’s record carries no scheduling data, and the docket’s recent filings ran through August 31 on contested discovery — evidence-preservation disputes, deposition fights, sealing motions — which is not the shape of a docket about to open. CourtListener/RECAP, which would carry the scheduling order itself, is bot-blocked even under headed rendering and is now recorded as a known-blocked surface. Verified and carried instead: Stability AI’s $76M Series B (August 25, 2026), with Electronic Arts and all three major music groups taking equity in the company they are litigating against.
Changelog trim: August 9, 2026 block removed from the main page inline (retained in the archive). Main page inline now shows exactly four blocks: September 1 / August 18 / August 16 / August 11.
Considered and excluded: Wan 3.0’s wider aggregator sweep beyond the eleven named — the pattern is recorded once in the entry rather than enumerated per platform. Pika Audio Models (Soundtrack, SFX, Speech, Music) — an audio model line, excluded per this page’s AI-video-first scope, same treatment as MiniMax Music 3. Higgsfield KÖK BÖRÜ at SIGGRAPH and the Global Film Festival thread — business and culture rather than spec or capability, the treatment already applied to Hell Grind and The Cully Hill Boys. “Cinema Studio 4” (Higgsfield) — still flagged, still not added: a second credit on an August 17 short strengthens the case it is a real named tool, but what changed from prior versions remains unconfirmed. EU Article 50(2) watermarking primary pin — still outstanding; the lane entry cites a page that carries no December date, flagged in place and not silently repaired here.

August 18, 2026

CORRECTION — Grok Imagine Video 1.5’s 1080p ship date was July 31, not August 1: The page stated August 1, 2026 in five places (Quick Verdict, Quick Reference row, spec card, What’s Next, and the closing prose). The correct date is July 31, 2026, per SpaceXAI’s own release page, whose published metadata reads July 31 in UTC. The error came from anchoring the date to a same-week amplification post on the @imagine sub-account (August 1, 20:32 UTC) rather than to the vendor’s own release page or its primary @grok announcement (August 1, 00:46 UTC — July 31 in both US Eastern and the company’s home Pacific time). All five occurrences corrected and the closing prose re-anchored to the release page. The substance is unchanged: 1080p is generally available, the feature set is as reported, and the $0.08/sec rate carries no resolution tier. Earlier changelog entries are left exactly as written per this page’s audit-trail contract; this entry is the correction of record. The affected roundup carries its own correction note — see Weekly Roundup — August 10, 2026.
MAGI-2 Preview — Sand.ai (NEW ENTRY, Open-Source & Local Generation): Verified and added a full entry for the August 5 open-weights release — 114B-parameter MagiMoE, ~6B active/token, 10-second clips with native audio at up to 1080p via a two-stage preview+refiner pipeline; ~307GB checkpoint; eight Hopper-class GPUs documented for inference; weights-only, no hosted API. License is unmodified Apache 2.0 with no territory carve-out — the direct contrast to MiniMax H3’s US/EU/UK/South-Korea-excluded Community License, checked three independent ways (repository LICENSE, repo root listing, model-card metadata). Debuted #6 on Artificial Analysis’s image-to-video leaderboard (Elo 1,104, fetched live 2026-08-18). Vendor: Sand.ai (Beijing), founded by Cao Yue (Swin Transformer co-author); MAGI-1 was its prior release (April 2025), and its bare unattributed bullet in Other Notable Open-Source Models is superseded by this entry and removed. Funding figures circulating in secondary coverage are not carried — no primary confirmation located. Closes the item queued in the August 16 “Considered and excluded” note.
Black Forest Labs — three-aggregator threshold reviewed, watch-entrant status held: The August 18, 2026 editorial review read three aggregator integrations in eight days (Runway August 4, Pika August 5, Luma August 12) plus a #2 Design Arena text-to-video debut as a pattern rather than distribution noise. What that pattern means — a commercial frontier lab, or the infrastructure underneath the platforms absorbing it — is the subject of a scheduled analysis. Big Nine membership stays open until that piece answers it; the What’s Coming entry now says so explicitly instead of carrying the question as an unresolved review note.
Sourcing rule applied going forward: a vendor ship date anchors to the vendor’s own dated release page where one exists, or to the primary announcing account — never to a downstream amplification post whose timestamp happens to be what got read.
Changelog trim: August 2, 2026 block removed from the main page inline (retained here). Main page inline now shows exactly four blocks: August 18 / August 16 / August 11 / August 9.

August 16, 2026

Weekly Roundup R#17 paired refresh: lastmod → 2026-08-16 (lead-by-one before the August 17 Roundup); params.toml ogImage → roundup-2026-08-17.png; Mode B infographic regenerated. Ships Sunday per the standard Stack-paired-with-Roundup cadence.
LTX — vendor identity corrected, entry moved to LTX-2.5: LTX has been an independent company since July 8, 2026, spun out of Lightricks (which kept its consumer apps, e.g. Facetune); CEO Zeev Farbman came with the video line. The entry header, the Quick Reference row, both Quick Verdict references, and the FAQ all read “LTX-2.3 (Lightricks)” and are corrected. LTX-2.5 shipped August 11 — Diffusion Fidelity Rendering, native multishot generation holding continuity across cuts, a custom Gemma 4 text-encoder backbone, HDR ACES support, open weights on Hugging Face, Day-0 ComfyUI support, and a same-window Runway integration. The 22M+ cumulative Hugging Face downloads figure since base LTX-2 in January is LTX’s own claim and is labeled as such. The spec bullets deliberately remain the verified LTX-2.3 card: the 2.5 launch post publishes no resolution, duration, or frame-rate numbers, so the card carries an explicit pending-verification note rather than letting 2.3’s figures inherit a 2.5 header.
HeyGen HyperFrames — MCP server inside Claude Design (August 7): animate a design, hit share, and a HyperFrames agent scores it, adds sound effects, tightens the motion, and returns a finished MP4 — no human in the step between design and shippable video. Added to the HyperFrames entry as the agentic-assembly pattern reaching a design surface rather than a video platform.
Black Forest Labs — third aggregator, plus a leaderboard debut: Luma integrated FLUX 3 Video on August 12, making three aggregators in eight days after Runway (Aug 4) and Pika (Aug 5). FLUX 3 Video also debuted #2 on Design Arena’s text-to-video board at Elo 1,496 — marked preliminary by Design Arena itself, a qualifier carried into the text. What’s Coming item updated; watch-entrant status explicitly held, with the three-aggregator threshold named as a live editorial question rather than silently treated as settled.
Seedance 2.5 — ComfyUI partner nodes, and the first resolution figure: ComfyUI shipped native Seedance 2.5 partner-node support August 7 (50 reference assets per run — 30 images, 10 videos, 10 audio; second-level shot control; dialogue and lip sync in 10+ languages), putting the model on the self-hosted layer. On August 14 Higgsfield opened free 1080p 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. Still no published consumer or developer price from any aggregator.
Changelog trim: July 26, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: August 16 / August 11 / August 9 / August 2.
Considered and excluded: Sand.ai’s MAGI-2 Preview — a 114B mixture-of-experts open-weights model under Apache 2.0 with no territory carve-outs (released August 5, ~307GB checkpoint, eight NVIDIA Hopper GPUs documented, Artificial Analysis image-to-video debut at #6), surfaced by the R#17 Cowork exit test. Real, on-beat, and a genuine gap against MiniMax H3’s excluded-territory license — but out of R#17’s coverage window and not yet re-verified to primary by the Writer-Researcher. Queued in next-agenda.md for a verified Stack row rather than added on a comparison document’s word. “Cinema Studio 4” (Higgsfield) — a production-tool version credited on the open-sourced ONEIRIC short and never logged here; flagged, not added, pending confirmation of what actually changed. ONEIRIC itself — a promotional open-sourcing, business/culture rather than spec or capability, same treatment already applied to Hell Grind and The Cully Hill Boys.

August 11, 2026

CORRECTION — California AI Transparency Act misidentified as SB 1000: The “State synthetic-media provenance mandates” item and the July 26 regulatory-timing note both attributed the August 2, 2026 effective date to California SB 1000. That was wrong. The operative statute is the California AI Transparency Act, SB 942 (2023–2024, Chapter 291), as amended by AB 853, which moved the operative date from January 1 to August 2. SB 1000 (2025–2026) is a separate, still-pending bill sharing the same popular name — passed the Senate 33–1 on May 19 with an urgency clause, still on the Assembly Third Reading File, never enrolled or signed. The provenance-mandates item now states the operative statute, its actual obligation and penalty (§ 22757.4), and lists SB 1000 and AB 2713 correctly as pending amending bills. AB 2713 was also clarified: its title is “California AI Transparency Act: system provenance data” and it amends § 22757.3.1 — it is a bill amending the Act, not the Act itself, which the prior wording blurred. All statuses verified against leginfo.ca.gov primary records. Earlier changelog entries are left exactly as written per this page’s audit-trail contract; this entry is the correction of record. Affected roundups (June 15 through August 3) corrected separately; see the correction note on Weekly Roundup — August 3, 2026.
Changelog trim: July 19, 2026 block removed from the main page inline (already in the archive). Maintains the four-most-recent inline window.

August 9, 2026

Weekly Roundup R#16 paired refresh: lastmod → 2026-08-09 (lead-by-one before the August 10 Roundup); params.toml ogImage → roundup-2026-08-10.png; Mode B infographic regenerated. Ships Sunday per the standard Stack-paired-with-Roundup cadence.
MiniMax H3 — open-weights fine print added (Cowork exit-test catch): weights shipped August 3 as promised, but the Community License excludes the US, EU, UK, and South Korea from use entirely, and local self-hosted generation caps at 768p, not the API’s 2K — both confirmed directly against the Hugging Face LICENSE file and model card. Updated the Weights bullet, added a new Caution bullet, and rewrote the closing prose paragraph.
Grok Imagine — 1080p ceiling closed: Grok Imagine Video 1.5 shipped native 1080p, voice references, and text-to-video on August 1 — the resolution gap this page has tracked since Musk’s missed April commitment. Updated the spec card, Quick Verdict, Quick Reference row, and prose; retired the “Grok Imagine Pro” framing since no separate tier was ever named again after the original announcement.
Seedance 2.5 — three aggregators shipped it the same morning: Runway, Higgsfield, and Pika all went live August 7, within about two hours of each other (06:11–07:59 UTC) — the aggregator race resolved. What’s Coming item rewritten; still no published consumer/developer price from any of the three.
Black Forest Labs — the pivot shipped: FLUX 3 (July 23) is the video model Scorsese’s June advisory role implied was coming — up to 20 seconds with native audio, Early Access only. Runway and Pika integrated it within 48 hours. What’s Coming item rewritten from “video model in development” to the actual launch; still a watch entrant, not a Big Nine row, pending a second release confirming the pivot is durable.
Changelog trim: July 14, 2026 block removed from the main page inline (already in the archive, below). Main page inline now shows exactly four blocks: August 9 / August 2 / July 26 / July 19.
Considered and excluded: Higgsfield’s Global Film Festival ($1M prize, Edwin Catmull juror) and Hell Grind’s full open-sourcing — real news (R#16 covers both), but business/culture events rather than spec or capability changes; no Stack row or spec-card change warranted. Cowork’s Seedance 2.5 Aug 7 pricing claim — sourcing too weak on both sides (search-summary only, no primary rate card located), not added; watch for R#17. Cowork’s “Grok Imagine Pro still 14 weeks overdue” — checked against this week’s actual 1080p ship; Cowork’s own note admits the figure is stale unrecomputed arithmetic, not a fresh finding — not adopted.

August 2, 2026

Weekly Roundup R#15 paired refresh: lastmod → 2026-08-02 (lead-by-one before the August 3 Roundup); params.toml ogImage → roundup-2026-08-03.png; Mode B infographic regenerated. Ships Sunday per the standard Stack-paired-with-Roundup cadence.
xAI → SpaceXAI vendor-identity change (WR-verified): SpaceX’s all-stock acquisition of xAI closed February 2, 2026 at a combined $1.25T valuation — CNBC’s Feb 3 report called it the biggest merger of all time; corroborated by Bloomberg (Feb 2) and NYT (Feb 2). The company dropped the xAI name on X, July 6: “We are now @SpaceXAI.” Grok and Grok Imagine keep their names — model IDs, pricing, and docs.x.ai URLs unchanged; Artificial Analysis already credits the model’s maker as SpaceXAI. Updated: entry header, summary-table vendor cell, the two forward-pointing “release notes” references, and the Grok Imagine Pro / 30-minute-video What’s Coming items. Historical narrative paragraphs (Jan–June 2026) kept the contemporaneous “xAI” name. Held out of the July 26/27 paired refresh pending verification (Carlos call 2026-07-26).
MiniMax H3 — MiniMax (NEW ENTRY, Big Eight → Big Nine): Added a full commercial-tier entry for the July 31 launch — successor to the Hailuo 2.x family; omni-modal input (text/image/video/audio); 5–15 sec clips at 2K (2560×1440), 24fps, native stereo audio; day-one API at $0.13/sec (model ID MiniMax-H3); consumer access via the Hailuo AI app; arena debut #2 audio T2V (Elo 1,242), #3 audio I2V (1,184), and #1 on Artificial Analysis’s video-editing board (1,130, 5,100+ samples, EIC-verified live) — the strongest non-Google arena debut this page tracks; weights promised “within days” under the MiniMax Community License (<$20M revenue, attribution) — flagged as a pledge, not a release, held to the same bar as HappyHorse’s open-weights claim; caution noted on unsigned EU-code status. Added a matching Quick Reference row and a new Quick Verdict “best for generative video editing” line. “Big Eight” renamed to “Big Nine” throughout (section header + two live cross-references in the Gemini Omni Flash entry; historical changelog blocks left as written, since they’re dated records).
EU Article 50 + transparency Code — resolved from “days out” to in effect: Article 50 took effect August 2 as scheduled. The Commission’s initial signatory list (published July 31, ~190 organizations) confirms Google and Meta (signed July 28) as the only AI-video-relevant signatories — ByteDance, Kuaishou, Alibaba, TikTok, SpaceXAI, and MiniMax all unsigned as enforcement begins. Added Article 50’s own penalty tier — no schedule of its own, falls under AI Act Article 99 at up to €15M or 3% of global turnover — surfaced by the Cowork blind exit-test and independently re-verified against the primary. Both What’s Coming items rewritten to current state.
Changelog trim: July 12, 2026 block removed from the main page inline (already in this archive, below). Main page inline now shows exactly four blocks: August 2 / July 26 / July 19 / July 14.
Considered and excluded: Cowork’s Seedance 2.5 framing around Jimeng/Doubao (China-domestic surfaces, secondary-sourced) — RCTV’s Dreamina-global-launch framing (primary-sourced) kept as written, reconciled as two brand surfaces of the same rollout rather than a contradiction; full reasoning in the R#15 draft’s EIC integration notes. Gemini Omni Flash’s I2V delist/relist arc (Cowork Story 3) — checked live, consistent with Cowork’s read, but benchmark churn on a board this page doesn’t treat as its own storyline; no Stack rewrite this pass.

July 26, 2026

Weekly Roundup R#14 paired refresh: lastmod → 2026-07-26 (lead-by-one before the July 27 Roundup); params.toml ogImage → roundup-2026-07-27.png; Mode B infographic regenerated. Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Veo — Performance Control added: new July 2026 technique (3D geometry/motion extraction from a real performer, independently editable player/environment layers), debuted in Google DeepMind’s Pelé “Gol da Rua Javari” reconstruction. Added to the Veo spec card + prose.
Runway — Media Router added: July 23 preference-optimized router (define “best” once — cost / quality / latency — auto-selects the model per request; live in Runway Dev). Added to Runway prose, extending the aggregator-to-router thesis.
Runway Agent — Physion-Arc #1 (hedged): a brand-new third-party agent benchmark ranked Runway Agent 2.0 #1 of 6; added to the Runway Agent entry with full caveat load (unreachable methodology page, primaries disagree on scale, Runway is the loudest amplifier, no independent replication).
Higgsfield — funding line added: reportedly in talks for $300–500M at a $5B valuation (~4× its Jan mark), ~$500M ARR (up from ~$200M end-2025, ~70% enterprise), DST Global reportedly in; round not closed. Added to the Higgsfield Agentic Layer entry, heavily hedged.
Regulatory timing — EU Art. 50 + CA SB 1000 now days out: Aug 2 effective date is six days out as of this refresh; EU Commission adopted final Article 50 guidance July 20. EU Art. 50 What’s Coming item updated; the EU transparency Code of Practice item’s July 22 signatory deadline noted as passed (roster pending).
Changelog trim: July 5, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: July 26 / July 19 / July 14 / July 12.
Considered and excluded: HeyGen “Companion Mode” for HyperFrames (July 22) — carried to R#15 prep, and a feature on the existing HyperFrames agent, not a new agent row; no [landscape] count change. xAI→SpaceXAI vendor-identity change — verified real (CNBC Feb 3 + live AA board relabel) but out of window and too large for a rushed paired refresh; held for a WR-verified next landscape pass (Carlos call 2026-07-26; filed in next-agenda.md). “Omni delisted from both AA I2V boards” (Cowork Story 2) — contradicted by a live AA fetch this session (Omni still #1 I2V, Elo 1,200); no change to the four-board-sweep line.

July 19, 2026

Weekly Roundup R#13 paired refresh: lastmod → 2026-07-19 (lead-by-one before the July 20 Roundup); params.toml ogImage → roundup-2026-07-20.png; Mode B infographic regenerated. Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Gemini Omni Flash — clean four-board sweep (corrects the no-audio boards): Omni took #1 on both Artificial Analysis no-audio boards (T2V Elo 1,327, I2V 1,374) on top of its with-audio #1, ending HappyHorse-1.0’s no-audio reign — the full re-rank deferred from July 14. HappyHorse-1.0/1.1 now sit #2/#3 no-audio; Seedance 2.0 #4 (T2V no-audio) / #2 (with-audio). Propagated to the Gemini Omni, Seedance, and HappyHorse entries, Quick Verdict, Quick Reference, and the above-the-fold summary.
Seedance 2.5 — BytePlus ModelArk API opened July 16: the 30-second single-pass model is now publicly callable (one call → one continuous clip + co-generated audio, up to 50 reference inputs), still with no published price; What’s Coming item + Seedance Note updated.
Seedance pricing reconciliation (deferred R#12 item resolved): the Stack’s AA-attributed ~$0.151/sec standard-tier figure reconciled against fal.ai’s live ~$0.30/sec 720p-with-audio rate — the estimate and the live commercial rate differ; the live gateway figure flagged as current for commercial pricing.
Kling — scale + Cannes added to the entry: 100M cumulative users across 224 countries; ARR ~$500M as of March 2026 (from ~$240M in Dec 2025; ~70–75% overseas). At Cannes Lions 2026 the inaugural “AI Craft” Grand Prix went to Google, not Kling (two Kling-made ad films won Silver + Bronze).
Arizona SB 1786 — corrected to “Dead”: recalled to the House and never enacted (Senate returned it May 4); azleg.gov Bill Status Inquiry lists it “Engrossed — Dead.” It died in the legislature; it was not vetoed.
NEW What’s Coming — PixVerse: Singapore lab closed a $439M Series C extension July 13 (valuation >$2B, Alibaba among investors); shipping the R1 real-time world model + a new V-Series; V6 sits mid-table on the AA audio board. Watch entrant, not a counted Big Eight row.
Changelog trim: June 28, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: July 19 / July 14 / July 12 / July 5.
Considered and excluded: Netflix’s ~300-titles disclosure (Netflix is a user of generative AI, not a model or lab — a Roundup story, not a Stack row); a PixVerse Big Eight row (V6 is mid-table with no category-leading axis — watch entrant only).

July 14, 2026

AI-search-optimization pass (extraction tuning, not a content refresh): added an above-the-fold “The short answer” summary — a self-contained, liftable one-paragraph answer to the head query (“best AI video model in 2026, how to choose”), one named-entity + checkable-fact verdict per axis, structured for LLM extraction and citation. Carlos-approved structural pass; scope in drafts/2026-07-14-stack-ai-search-pass-brief.md.
Freshness stamps: Quick Verdict and Quick Reference now carry explicit “as of [date]” / board-source captions (the trait-checklist freshness signal). Fixed the intro “Last updated” prose, which read July 5 while lastmod was July 12 — the freshness-drift class this pass exists to catch.
Section extraction audit: the spec-card-first structure (Option A, June 11) is confirmed extraction-ready — no structural change to the Big Eight, Agentic Layer, or routing-framework sections; volatile facts already sit in canonical spec-card / table cells a model can lift.
Paired (separate files): the grok-imagine-vs-seedance-2-0 comparison page refreshed to current specs + citation-trait structure (the recurring-comparison-cadence pilot); per-model tag-page funnels spec’d for operator application (Hugo template, owned territory).
Changelog trim: June 21, 2026 block removed from the main page inline (already in the archive). Main page inline now shows exactly four blocks: July 14 / July 12 / July 5 / June 28.
lastmod: 2026-07-12 → 2026-07-14. params.toml [landscape] counts + ogImage unchanged (not a roundup pass; no infographic).
Same-day currency update (the extraction audit caught it live): Gemini Omni Flash took #1 on both Artificial Analysis with-audio boards July 14 (T2V Elo 1,240, I2V 1,203), passing Seedance 2.0 — new within 48h of R#12 reporting the AA score as pending; verified live + against the Grok-vs-Seedance refresh. Propagated to the Gemini Omni + Seedance + HappyHorse entries, Quick Verdict, Quick Reference, the above-the-fold summary, and the Omni Pro What’s Coming item. Deferred to the R#13 weekly refresh: the full with-audio board re-rank (Wan2.7 #3, HappyHorse positions) and the Seedance fal.ai-vs-AA pricing reconciliation ($0.3034/sec@720p-audio vs the Stack’s AA-attributed $0.151/sec).
Considered and excluded: rewriting the spec cards / routing framework (Option A already made them extraction-friendly — churn risks introducing errors for marginal gain); AEO visibility-score tooling (a thermometer, not a lever); genericizing the voice for “SEO” (the non-commodity operator voice IS the citation asset, not an obstacle to it).

July 12, 2026

Last updated date: Advanced from July 5 to July 12, 2026. lastmod set to 7/12 deliberately so the July 13 Weekly Roundup (R#12) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Kling — record round closed: Kuaishou disclosed July 2 a ~$3B round at an ~$18B post-money valuation (Tencent, Alibaba Cloud, Baidu among 38+ investors; Kuaishou 68.33% stake; Hong Kong IPO framing; ~$500M ARR, +300% YoY). Replaced the May “preliminary financing” paragraph in the Kling entry; the round priced below the ~$20B May target.
Kling MCP & CLI (July 8): Kling’s model is now agent-callable (MCP endpoint + CLI) — added to the Kling spec card. No new Agentic Layer row: Kling exposes a model, it isn’t an orchestrator.
Runway Dev (July 8): dev/enterprise platform — one API over Runway’s own models (Gen-4.5, Aleph 2.0, Act-Two) plus third-party (Seedance, GPT Image 2, ElevenLabs); Recipes + Workflows; Adobe/ElevenLabs/Shutterstock/Figma Weave/Gamma/Silverside customers; SOC 2 Type II + IP indemnification. Added to Runway prose.
Pika Director’s Suite (July 9): AI timeline editor for agent-driven video + VFX / Seedance-4K-VFX Skills. Added to Pika prose.
Higgsfield NLE plugins (July 3): Gemini Omni Flash + Seed Audio 1.0 into Adobe Premiere Pro + DaVinci Resolve timelines. Added to the Higgsfield Agentic Layer entry.
HeyGen HyperFrames — “HTML Is All Agents Need”: James Russo X-Article on HTML-native agent-video rendering (1.3M videos / 267K creators / 32K stars). Added to the HeyGen Agentic Layer entry.
Grok Imagine — single-request 15-second takes (July, user-reported): added to the Grok spec card and the What’s Coming Grok item — hedged (no xAI release note; the 15-second ceiling was already reachable via chained extensions).
Wan 2.7 — Wan2.7-260612 benchmark: dated checkpoint debuted #2 on Artificial Analysis text-to-video-with-audio (Elo ~1,160). Added to the Wan spec card.
Seedance 2.5 — rollout began July 3: What’s Coming item updated from “enterprise beta” to public rollout (Dreamina/Jimeng live, CapCut mid-July, API late July; vendor-claimed, no US/API date).
NEW What’s Coming — Meta Muse Video: MSL previewed July 7 (first in-house Meta video model, #3 on Design Arena, preview-not-product); watch entrant, not a counted Big Eight row.
NEW What’s Coming — New York Synthetic Performer Disclosure Law: live June 9; ad-specific AI-disclosure mandate; compliance-staircase addition (the out-of-window catch home for R#12).
Changelog trim: June 14, 2026 block removed from the main page inline section (already in the archive). Main page inline now shows exactly four blocks: July 12 / July 5 / June 28 / June 21.
params.toml: ogImage → roundup-2026-07-13.png. [landscape] counts unchanged (commercial 8 / open_source 10 / agentic 6 — Meta Muse is a watch entrant; Kling MCP adds no orchestrator row).
Considered and excluded: TwelveLabs $100M Series B (video understanding, off-beat); Meta Muse Image / Seedream 5.0 Pro (image models). Big Eight commercial count unchanged — Muse Video is a preview with no specs.

July 5, 2026

Last updated date: Advanced from June 28 to July 5, 2026. lastmod set to 7/5 deliberately so the July 6 Weekly Roundup (R#11) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Gemini Omni Flash — public preview, not GA (June 30, 2026): Corrected the access-state framing from the May I/O-preview claim to the actual shipped spec: public preview via Google AI Studio, the Gemini API, and Gemini Enterprise Agent Platform (also live in the Gemini app and Google Flow); $0.10/sec; 10-second cap (“longer durations coming soon”); 720p native — no 1080p or 4K tier, independently corroborated by VentureBeat against Google’s own launch post (the brief’s original “1080p consumer” tier did not hold up under primary verification and was dropped); conversational multi-turn editing via the stateful Interactions API; SynthID + C2PA Content Credentials by default. Added the July 2 Design Arena #1 debut (Elo 1,404, +101-point margin) — flagged explicitly as Design Arena, not Artificial Analysis (the board tracked for the rest of the Big Eight); AA placement still pending; Seedance 2.0 still leads Design Arena’s own image-to-video board. Updated spec card, prose, Quick Verdict bullet, Quick Reference row, and the “Gemini Omni Pro shipping window” What’s Coming item.
Higgsfield Supercomputer: Added MCP audio generation (June 27 — Seed Audio 1.0 + ElevenLabs v3, 50+ languages), Explainer (July 2 — faceless documentaries up to 10 min, Claude Fable 5 + Gemini Omni Flash), and Higgsfield making its MCP free (July 2 — 100 credits, 3-day full-access trial). Updated Agentic Layer table row and Higgsfield prose.
Runway Agent: Added the Bertelsmann partnership (July 1 — RTL Group, BMG, Bertelsmann Marketing Services; a genuine enterprise-adoption story, not a pilot) and Agent Skills (July 2 — slash-command layer on Runway Agent for ad campaigns, commercials, localized ads). Runway is now the third vendor, after Pika and Luma, to converge on “Skills” as the term for portable agent workflows. Updated Agentic Layer table row and Runway Agent prose.
NEW — HeyGen HyperFrames added to The Agentic Layer: New row and entry for HeyGen’s HyperFrames “Next Generation Skills” (launched June 26, 2026) — agentic routing across 9 production workflows (launch video, music video, captions/overlays, +6 more), context-based routing with no manual workflow selection; 1.53M impressions, the highest-engagement AI-video product launch RCTV tracked in the June window; single-vendor (HeyGen’s own models). Surfaces at hyperframes.heygen.com. Agentic Layer intro and the “Agentic-orchestration layer” What’s Coming item updated from five agents to six.
HappyHorse-1.1 — settled #2 on both Artificial Analysis with-audio boards: T2V with-audio: Seedance 2.0 #1 (Elo 1,222), HappyHorse-1.1 #2 (Elo 1,151), HappyHorse-1.0 #3 (Elo 1,125). I2V with-audio: Seedance 2.0 #1 (Elo 1,194), HappyHorse-1.1 #2 (Elo 1,116), HappyHorse-1.0 #5 (Elo 1,089 — dropped two spots as Grok Imagine Video 1.5 and Wan 2.7 climbed above it). Seedance 2.0 remains #1 on both. Open-weights delivery for HappyHorse-1.0 remains unshipped (69 days past the April 27 API launch). Updated HappyHorse spec card, prose, and Quick Reference row; Seedance Note bullet’s T2V-audio Elo also reconciled (1,219 → 1,222).
What’s Coming — Seedance 2.5 refreshed: Added joint audio-video generation and “no pricing announced” to the spec list; public launch remains targeted for early July 2026, enterprise beta still open (unshipped as of this update).
What’s Coming — EU transparency Code of Practice added: The EU AI Office’s marking-and-labelling Code of Practice (published June 10, 2026) opens for signatures with a July 22, 2026 deadline, ahead of the binding Article 50 disclosure date (August 2).
Changelog trim: June 11, 2026 block removed from main page inline section (already in this archive). Main page inline now shows exactly four blocks: July 5 / June 28 / June 21 / June 14.
params.toml [landscape]: Bumped agentic = 5 → agentic = 6 (HeyGen HyperFrames row). commercial = 8 and open_source = 10 unchanged.
Considered and excluded: A separate Big Eight row for HeyGen (it’s an agentic orchestrator over HeyGen’s own models, not a new base video model — same test applied to Higgsfield and Runway Agent); no change to the Quick Verdict order.

June 28, 2026

Last updated date: Advanced from June 21 to June 28, 2026. lastmod set to 6/28 deliberately so the June 29 Weekly Roundup (R#10) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
HappyHorse-1.1 enters the Artificial Analysis leaderboard: First model update from Alibaba ATH’s AI Innovation Unit since the 1.0 debut in April. T2V no-audio: HappyHorse-1.0 remains #1 (Elo 1,290, revised from 1,293); HappyHorse-1.1 debuts at #2 (Elo 1,285); Seedance 2.0 falls to #3 (1,274). I2V audio: Seedance 2.0 #1 (1,194), HappyHorse-1.1 #2 (1,119), HappyHorse-1.0 #3 (1,089). Audio T2V: Seedance 2.0 #1 (1,219). Updated Quick Verdict, Quick Reference table, HappyHorse spec card (Benchmark position bullet), and HappyHorse prose (new HH-1.1 paragraph). No separate 1.1 pricing or endpoint announced; open-weights for 1.0 remain undelivered at week 9. Source: Artificial Analysis T2V leaderboard, Artificial Analysis I2V leaderboard (verified June 28, 2026 — cited in R#10 sourcing).
Seedance 2.0 — 4K native confirmed at FORCE: At Volcano Engine FORCE (June 23, 2026), ByteDance confirmed Seedance 2.0 now supports native 4K with 10-bit color depth. Seedance spec card Max resolution updated from 2K to 4K native, 10-bit color depth; Seedance prose adds FORCE note. Benchmark Note bullet updated: audio T2V Elo 1,215 → 1,219; T2V no-audio position #2 → #3 (behind HH-1.0 and HH-1.1). Source: The Decoder — ByteDance’s Seedance 2.5 breaks the 30-second barrier.
Luma — Timeline editor / EDL Export (June 19) + Luma Connectors (June 24): Two Luma AI feature updates that cleared the roundup’s scope gate but belong in the Stack. Timeline editor adds multi-clip editing with Edit Decision List export for NLE handoff. Luma Connectors adds native Airtable, Dropbox, and Google Drive integrations. Both added as spec card bullets in the Luma Ray 3.2 / Ray 3.14 entry.
What’s Coming — Seedance 2.5 added: ByteDance previewed Seedance 2.5 at FORCE (June 23): 30-second native clips with no post-stitching, up to 50 multimodal reference inputs, targeted post-generation editing. Enterprise beta now; public launch early July. All specs vendor-claimed; no independent benchmarks yet. Added before Dreamina Octo in the ByteDance cluster.
What’s Coming — California AB 2713 added: Added AB 2713 (California AI Transparency Act) to the State synthetic-media provenance mandates item as a parallel California track distinct from SB 1000.
params.toml ogImage: Updated from roundup-2026-06-22.png to roundup-2026-06-29.png. Landscape counts unchanged: commercial 8 / open_source 10 / agentic 5 — no new rows this week.
Considered and excluded: Kling 3.0 Turbo pricing details via third-party resellers (reseller rates belong in roundup; Stack waits for Kuaishou official pricing); HappyHorse-1.1 as a separate Big Eight row (model-update-in-place, same API surfaces).

June 21, 2026

Last updated date: Advanced from June 14 to June 21, 2026. lastmod set to 6/21 deliberately so the June 22 Weekly Roundup (R#9) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Luma — Ray 3.2 (June 9, 2026): Section renamed from “Luma Ray 3.14” to “Luma Ray 3.2 / Ray 3.14.” Ray 3.2 and Ray 3.14 are parallel sub-models in the Ray3 family — not sequential versions. Ray 3.2 (June 9) adds up to 16 keyframes per clip, performance tracking across 8 faces, native HDR with 16-bit EXR export, Enhanced Reframe, clips up to 20 seconds at 1080p, and first API availability. Ray 3.14 remains available for duration-change and loop workflows where its fixed-length output is an asset. Spec card, prose, Luma Agents table, and all Routing Framework deep-links updated. Source: Luma blog — introducing-ray-3-2, Luma learning center — ray-3-2-introduction-and-core-concepts.
Grok Imagine — Video 1.5 GA (June 16, 2026): Updated from “shipped as public API preview June 3” to general availability June 16 across Imagine API, grok.com, iOS, and Android. Added Video 1.5 Fast variant (~25s for a 6-sec 720p clip). Pricing reconciled: original Grok Imagine is $0.05/sec ($3.00/min, no native audio); Video 1.5 GA is $0.08/sec ($4.80/min, audio-native) — single published rate, no resolution tier on any xAI docs page. The “$4.20/min” figure that was in the Quick Verdict pointed to the old $0.07/sec@720p rate of the original model — corrected. Elo updated to 1,113 (from 1,110) per June 2026 AA leaderboard. Source: docs.x.ai/developers/models (re-verified June 21).
Kling — Turbo and O3 (June 17–18, 2026): Added Kling 3.0 Turbo (speed/cost tier: faster generation, lower price, improved lip-sync, stable motion) and Kling 3.0 O3 (up to 15s at full 4K, stronger prompt/reference consistency). Rolled to 7 partner platforms simultaneously: fal, SeaArt, Clipfly, Fotor, GlamAI, Runware, Morphic. Kuaishou published no official pricing; capability claims attributed to partner-platform announcements (independently corroborated across all 7 platforms). klingai.com remained 446-blocked. Source: fal.ai/kling-3 + 6 partner confirmations.
Seedance — two new access channels (June 16–17, 2026): Added Higgsfield “Seedance Unlimited” (June 17) — exclusive non-ByteDance unlimited-Seedance surface via BytePlus (ByteDance’s B2B cloud), “Enhanced Seedance 2.0 Fast” model, 30-day add-on tiers, 480p–720p. Added Dreamina Seedance 2.0 mini (June 16) — emerging markets (SE Asia, ME, Africa, Europe, SA; US excluded) at ~$0.02/sec, roughly 7× below standard 720p tier. Both added to spec card and Seedance prose. Source: @higgsfield X post June 17 (primary, xservice); @dreamina_ai X post June 16 (primary, xservice).
What’s Coming — Grok 1080p item updated: Noted Video 1.5 GA while preserving the 1080p-Pro-still-missing tracking.
Changelog trim: May 31, 2026 block removed from main page inline section; moved to archive (already present). May 25, 2026 block confirmed in archive. Main page inline now shows exactly four blocks: June 21 / June 14 / June 11 / June 7.
Considered and excluded: No new Big Eight rows (Ray 3.2 is an update to the Luma entry; Kling Turbo/O3 are variants of the existing Kling 3.0 entry; Seedance mini and Higgsfield Unlimited are access-channel additions, not new Big Eight entries). params.toml landscape counts unchanged: commercial 8 / open_source 10 / agentic 5.

June 14, 2026

Last updated date: Advanced from June 11 to June 14, 2026. lastmod set to 6/14 deliberately so the June 15 Weekly Roundup (R#8) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
The Agentic Layer — Higgsfield Supercomputer: Added the early-June distribution sprint — five external surfaces in 12 days (Claude MCP May 28, Adobe plugins May 29, Figma June 4, Minecraft mod June 5, DaVinci Resolve plugin June 8) plus the Grok Imagine 1.5 integration into Higgsfield’s own platform — the clearest routing-bet expression yet (a competitor’s model inside Higgsfield’s surface).
Other Notable Open-Source — NVIDIA Cosmos 3: Updated the launch-time “claim not independently confirmed” to live-board confirmation — Cosmos3-Super leads open-weight image-to-video at Elo 1,251 (June 2026).
What’s Coming — two updates: Added Varya (Avataar.ai, June 11 — India IndiaAI-Mission open-weight Wan-2.2 distillation, vendor-reported ~$0.005/sec; watch entrant, specs unverified, not a tracked row). Rewrote the EU AI Act Article 50 item from a flat August cliff to the two-step deadline — disclosure August 2, machine-readable watermarking December 2 for in-market systems (May 2026 AI Omnibus grace period; surfaced by the R#8 Cowork exit-test).
Considered and excluded: Varya as a counted open-source row (specs Indian-outlet-only, AA-unbenchmarked, vendor page down — watch entrant until verified); the Cowork “HappyHorse weights flipped to published” claim (our render still showed “Coming Soon” — held out as unverified, HappyHorse open-weights item unchanged).
Paired update outside this file: params.toml ogImage → roundup-2026-06-15.png; [landscape] counts unchanged (commercial 8 / open_source 10 / agentic 5 — Varya is a watch entrant, not a counted row). Changelog trim: the May 25 block moves to the archive on the next pass to restore the four-most-recent display.
Process note: Stack deltas drafted blind of the Cowork comparison (read only after R#8 was built, exit-test discipline); the comparison’s one verified catch — the EU Art. 50(2) grace period — is reflected here and in the roundup’s What-to-Watch.

June 11, 2026

Page restructure (reference-card format): Every model and agent entry now leads with its spec card — the canonical location for volatile facts (pricing, resolution, arena positions, access) — followed by current-state analysis. Relative-time references removed page-wide (11 instances); duplicated stats reconciled to single canonical values (HappyHorse-1.0 T2V no-audio Elo confirmed at 1,293 against the live Artificial Analysis board). No facts removed: dated event history lives in this changelog.
Changelog split: The Stack page now carries the four most recent update blocks; this archive holds the full history and is always complete.
Navigation anchors: Stable anchors added to every model and agent entry; Quick Verdict bullets and Routing Framework answers now deep-link to entries.

June 7, 2026

Last updated date: Advanced from May 31 to June 7, 2026. lastmod set to 6/7 deliberately so the June 8 Weekly Roundup (R#7) leads as the headline article rather than tying with this Stack update (lead-by-one rule). Ships Sunday per the standard Stack-paired-with-Roundup cadence.
Tags: Added cosmos, black-forest-labs, flux, agnes-video — NVIDIA Cosmos 3 enters open-source tracking; BFL/FLUX and Agnes-Video surface as What’s-Coming entrants.
Grok Imagine — xAI: Added Grok Imagine Video 1.5 (June 3) — xAI’s first native image-to-video API ($0.08/sec, 60 RPM, audio output, alias grok-imagine-video-1.5-2026-05-30); opens #2 on the Artificial Analysis I2V with-audio arena (Elo 1,110), behind Seedance 2.0 and ahead of HappyHorse-1.0. xAI’s #1-debut claim not borne out by the live board. New model paragraph + Arena / Max-resolution bullets; Quick Verdict + Quick Reference Grok row updated.
Arena standings refresh (verified against live Artificial Analysis boards, June 7): T2V no-audio — HappyHorse-1.0 #1 (1,293), Seedance 2.0 #2 (1,274), Grok #5 (1,235). T2V with-audio — Seedance 2.0 #1 (1,215), HappyHorse #2 (1,122): a 92-point gap reopened after a June recalibration (near-tied a week earlier). I2V with-audio — Seedance 2.0 #1 (1,194), Grok 1.5 #2 (1,110), HappyHorse #3 (1,094). Corrected a stale board-conflation in the prior Quick Verdict / HappyHorse standings. Updated Quick Verdict (HappyHorse + Grok), Quick Reference (Grok + HappyHorse rows), and the Seedance / HappyHorse / Grok model-entry benchmark bullets.
Other Notable Open-Source — added NVIDIA Cosmos 3: Open-weights omnimodel (Cosmos 3 Super + Nano on Hugging Face, OpenMDW-1.1, June 1; 291-author arXiv paper). NVIDIA’s “top open-source T2I/I2V” claim attributed, not independently confirmed.
The Agentic Layer — Higgsfield Supercomputer: Backbone upgraded to Claude Opus 4.8 (May 29); Higgsfield Reframe (May 30, MCP-native) noted. Table cell + paragraph.
What’s Coming: Rewrote the “AI executive order” item from postponement to signed June 2 (voluntary 30-day frontier review, no-licensing language, cybersecurity package; OpenAI/Anthropic/Google welcomed) — resolves the May 21 pull. Added “Runway Cosmos Coalition” (June 1 alliance, base model unspecified), “Agnes-Video — a new API price floor” ($0.30/min, Elo 905 #24), and “Black Forest Labs enters the video conversation” (Scorsese adviser, FLUX.2, $300M Series B). Updated the HappyHorse open-weights item to ~6 weeks, AA-API-only.
Considered and excluded: Agnes-Video as a Big Eight row (Elo 905, price-floor not production-grade — What’s-Coming watch); NVIDIA Cosmos 3 as a Big Eight row (open-weights physical-AI omnimodel — Open-Source Notable); Luma Human After All / Open Physical AI Lab (partnership + research lab, no shipped model — roundup material, not Stack).
Paired update outside this file: params.toml [landscape] open_source 9 → 10 (Cosmos 3 added); commercial (8) and agentic (5) unchanged. R#7 (ai-video-weekly-roundup-2026-06-08.md) reads the Stack’s live .Lastmod for its up_next card — no date sync needed (template change shipped 5/31).
Process note: Phase 1 pre-flight surfaced the 11-item change list (all standings pre-verified against live AA boards) before this Phase-2 atomic Edit sequence; no full-file rewrite (transcription-risk avoidance, ~660-line page). Stack deltas drafted blind of the June-5 Cowork comparison drop, read only after R#7 was built (exit-test discipline); the comparison’s verified catches — EO signing, Agnes, the arena recalibration — are reflected here.

May 31, 2026

Last updated date: Advanced from May 25 to May 31, 2026. lastmod set to 5/31 deliberately so the June 1 Weekly Roundup (R#6) leads as the headline article rather than tying with this Stack update (lead-by-one rule); R#6 up_next.date synced to 5/31 in the same pass. Ships today (Sunday) per the standard Stack-paired-with-Roundup cadence.
NEW SECTION — The Agentic Layer: Orchestration on Top of the Models {#agentic}: Dedicated H2 placed after The Big Eight, before Open-Source. Resolves the open structural question raised 2026-05-18 (agentic layer had no home in the Stack). Build derived from drafts/2026-05-26-agentic-orchestration-brief.md (Carlos-approved); decision-to-build pulled forward from the 6/2 gate to today. Five confirmed agents in a 7-column orchestration table (Agent · Vendor · Orchestrates · Single/Multi-model · Surfaces · What it automates · Access) + per-agent paragraphs: Luma Agents, Pika Agents (PikaStream 1.0 as a sub-bullet — runtime, not orchestrator), Runway Agent, Higgsfield Supercomputer, Adobe Firefly AI Assistant. Lead paragraph carries the “model is the engine, agent is the steering wheel” thesis; what-to-watch footer names three axes (single/multi-vendor convergence, AI-native vs incumbent, agent-to-agent interop). Archive callbacks to R#2 (Pika Agents lede) and R#4 (Mazda/Boundless) threaded.
Census frontier closed: Krea AI and LTX Studio both evaluated against the agent-vs-aggregator test and excluded — Krea is a menu-driven model picker (REST API, no conversational orchestration); LTX Studio is a timeline/studio tool (multi-model but menu-driven, no persistent cross-session context). Both fit the deferred “Tools” framing, not the agentic section.
Intro repositioned: “AI video software stack” → “AI video software stack — models, orchestration agents, and open-source generation,” reflecting the new three-part structure.
Arena updates (routine refresh, through 5/29): HappyHorse-1.0 repositioned to #2 on Artificial Analysis T2V no-audio (Elo 1,212; Seedance 2.0 now #1 at 1,215 — 3-Elo gap, inside the noise band); #1 on I2V no-audio unchanged. Grok Imagine added at #5 T2V no-audio (Elo 1,234 — effectively tied with Kling 3.0 Omni at #4). Updated in the Quick Verdict, Quick Reference table, and the HappyHorse + Grok Imagine model entries; new Quick Verdict bullet “Best multi-model orchestration agent — Pika Agents.”
What’s Coming — three new items: Reactor (real-time world models as infrastructure — $59M stealth exit May 28, Lightspeed-led, Katzenberg board observer; a category the Stack doesn’t yet track); MCP as a distribution layer for AI video (Runway/Pika/Higgsfield MCP cluster May 27–28 + Higgsfield Adobe Premiere/After Effects plugins); Dreamina Octo (pre-launch, “Vibe Create,” early-access survey live). Existing agentic-orchestration item updated from four to five agents with #agentic anchor links and “now has its own dedicated section.”
Homepage band — third tier: params.toml [landscape] agentic = 5 added; layouts/index.html extended to a three-tier band (agentic count read, three-way cell loop, third legend swatch + counter card linking #agentic); static/css/rctv.css adds --landscape-agentic: #ece4d0, cell--agentic, swatch--agentic. The #ece4d0 reuse (same hex as the Mode B roundup-infographic background) is an acknowledged cross-surface exception, documented in knowledge/design-system-brief.md.
Paired updates outside this file: .claude/commands/rctv-landscape-update.md (§ Structure gains the Agentic Layer; § Models to Watch resolves the open structural question + adds a “Tracked agents” subsection with inclusion criteria; params.toml sync step now three-tier); knowledge/design-system-brief.md (new “Landscape band — tier colors” section); content/posts/ai-video-weekly-roundup-2026-06-01.md (up_next.date 5/25 → 5/31).
Process note: Agentic-section content + routine-refresh deltas drafted by the Writer-Researcher blind of the sealed Cowork landscape comparison (knowledge/transitions/2026-05-29-roundup-comparison/, held for a post-build exit-test); Hugo template / CSS / params / paired-knowledge edits by the EIC. Phase-2 atomic Edit sequence, no full-file rewrite (transcription-risk avoidance on the now-630-line page).

May 25, 2026

Last updated date: Advanced from May 24 to May 25, 2026. Off-cycle mid-week edit (the standard cadence is Stack-paired-with-Roundup on Mondays). Operator-approved out-of-band to capture an additional week of GSC indexing data before the next settled-week read
Quick Verdict section added: New ## Quick Verdict: Best AI Video Model by Use Case (2026) block inserted between the intro paragraph and the Quick Reference table. Nine declarative “Best for X — Model Y” verdicts each carrying a one-line judgment, anchored at #quick-verdict, with a section-level link forward to #how-to-choose-a-routing-framework for the deeper decision tree. Verdicts: photorealism (Veo 3.1), broadcast 4K (Kling 3.0), character consistency (Seedance 2.0 Pro), stylized/VFX (Runway Gen-4 Turbo), editing existing video (Aleph 2.0), free consumer distribution (Gemini Omni Flash), open-source local generation (LTX-2.3), benchmark quality with API (HappyHorse-1.0), cheapest commercial API (Grok Imagine). Designed to capture head-keyword queries (“best ai video generation models 2026 comparison,” “top text-to-video ai models comparison 2026”) that GSC W20–W21 data showed ranking at positions 70–95 despite the Stack containing the underlying content
Why this lands here, not in a new article: The Stack already covers the comparison data; the missing element was framing the data the way head-query searchers expect (declarative verdicts vs. a 499-line reference). Improving the Stack’s head-query SEO is Tier 1 of a three-tier plan; Tier 2 (selective “X vs Y” comparison pieces under the mid-tier Sonnet slot) and Tier 3 (standalone “Best AI Video Model 2026” article) are gated on the Tier-1 ranking lift over 4–6 weeks of crawl cycles
Considered and excluded: Direct anchor links to individual model entries (Hugo’s auto-generated heading anchors are version-specific and brittle for headings with slashes / em-dashes — chose section-level link instead); a “Best for production volume — Luma Ray 3.14” verdict (Luma is on the Big Eight but doesn’t dominate a unique head-query slot; would risk diluting the more recognizable verdicts); agentic-orchestration platforms in the verdict list (they’re orchestration layers, not single-model verdicts — stays in #how-to-choose-a-routing-framework, awaits the 2026-05-26 ops meeting structural decision)
Paired update outside this file: Synced up_next.date in content/posts/ai-video-weekly-roundup-2026-05-25.md from 2026-05-24 to 2026-05-25 so R#5’s teaser card matches the current Stack lastmod. Both surfaces now share the 2026-05-25 date; Roundup leads in time-sorted feeds via its T05:00:00 publish time vs. Stack’s default-midnight lastmod
Process note: Phase 1 pre-flight surfaced full verdict prose to operator before Phase 2 atomic-write; no per-edit pauses within Phase 2 (same precedent as 2026-05-18 / 2026-05-24)
Sora 2 — OpenAI entry moved to end of Big Eight section: Sora 2 is discontinued (consumer shutdown executed April 26, 2026; API sunset September 24, 2026) and previously sat in the first position immediately after the section intro promising “production-grade models dominating professional and creator workflows in early 2026” — an editorial tension between the intro’s dominance claim and a tombstone-as-first-entry. Move-to-end resolves it: active models lead the section (Luma Ray 3.14 now first), Sora 2 closes as the historical reference and post-mortem anchor. Entry content preserved verbatim per CLAUDE.md “never remove without good reason” discipline; only the position changed.
How to Choose routing framework restructured for SEO efficiency: All 19 bullets reworded from need-form (Need X? → Y) to declarative-question form anchored on head keywords (Which AI video model wins X? Y.). Targets long-tail question-shape queries Google’s NLP wasn’t surfacing us for under the prior Need X? framing. Phrasing varied across the list to avoid drone effect. Answer-on-same-line preserved so routing-decision scan time stays fast. Punctuation shifted from → (flowchart) to ? + . (prose).

May 24, 2026

Last updated date: Advanced from May 18 to May 24, 2026. The lastmod is set to 5/24 deliberately so the May 25, 2026 Weekly Roundup (R#5) leads as tomorrow’s headline article rather than being chronologically tied with this Stack update — operator-explicit instruction
Tags: Added omni, gemini, aleph — Gemini Omni launched at I/O 2026; Aleph 2.0 is Runway’s May 21 editing-model launch
The Big Seven → The Big Eight: Section heading updated; commercial entry count rises to 8 with Gemini Omni Flash joining
Quick Reference table: Added Gemini Omni Flash row (Best For: multimodal input, broadest free distribution, SynthID provenance; Max Resolution: not published at launch; Free Tier: YouTube Shorts + YouTube Create; Paid From: $19.99/mo Google AI Plus; API: coming weeks). Renamed Runway row from “Gen-4 Turbo / Gen-4.5” to “Gen-4 Turbo / Aleph 2.0” with multishot edit propagation added to Best For
Gemini Omni Flash — Google DeepMind: New commercial entry, placed after Veo 3.1. Launched May 19, 2026 at Google I/O — Flash tier shipped that day; Pro tier on “soon” with no shipping date. Multimodal any-to-any input (image / audio / video / text → video output); SynthID watermark by default on every output. Paid via Google AI Plus, Pro, and Ultra in the Gemini app and Google Flow; free distribution rolling into YouTube Shorts and YouTube Create; developer and enterprise APIs in “coming weeks.” Honest hedge on resolution and duration — Google’s launch blog published no specs. Internal links to the Omni flagship analysis (/posts/omni-world-model-claim/) and to the May 25 Weekly Roundup for the distribution news-beat
Runway Gen-4 Turbo → Runway Gen-4 Turbo / Aleph 2.0: Entry heading renamed. Added paragraph on Aleph 2.0 launch (May 21) — upgraded video editing model with single-frame edit propagation across the rest of a clip; multishot sequences up to 30s at 1080p edited across cuts in one pass; localized edits via reference frame; ships inside the new Edit Studio product surface; available on all paid Runway plans on desktop web. Updated Max resolution, Key features, Access, and Next spec bullets to reflect Aleph 2.0 and Edit Studio
How to Choose routing framework: Added “Need free AI video inside an app you already use? → Gemini Omni Flash via YouTube Shorts and YouTube Create”; added “Need multimodal input (image + audio + video → video)? → Gemini Omni”; added “Need to edit one frame and have it propagate across a 30-second multishot? → Runway Aleph 2.0 in Edit Studio”
What’s Coming — removed “Google I/O 2026 (May 19–20)”: Resolved. Omni launched and is now a Big Eight entry alongside Veo 3.1; Veo 4 was not announced at I/O (verified against the Pichai keynote primary)
What’s Coming — “White House FDA-style AI vetting EO” → “AI executive order revival posture”: On May 21, Trump postponed signing the FDA-style frontier-model review EO hours before the scheduled ceremony (Axios + PBS NewsHour primary verification); the framework that would have generalized CAISI’s bilateral pre-deployment evaluation agreements across the field is on hold. Whether the order returns in slimmer form within weeks (Hassett-side) or stays shelved indefinitely (Sacks-side) defines federal AI policy through the rest of 2026 — and the launch calculus for every commercial video model that would have landed inside the framework
What’s Coming — “TAKE IT DOWN Act platform compliance” → “TAKE IT DOWN Act enforcement first action”: Item rewritten to current state. May 19 deadline passed; FTC opened formal enforcement on schedule (TakeItDown.ftc.gov stood up, business-guidance documentation published); same-day second-wave warning letter to 12 “nudify” tool sites on top of the May 13 fifteen-platform cohort. First enforcement target — platform, tool, or generator — sets the operative precedent for the rest of the year
What’s Coming — “State synthetic-media provenance mandates” updated: Added California SB 1000 progression (Senate passed 33–1 with urgency clause May 19; urgency designation triggers immediate effect on signature) and Hawaii HB 2137 status (sat 18 days on Governor Green’s desk without signature; capitol.hawaii.gov returns 403 to automated fetches; status verified against Transparency Coalition tracker). Arizona SB 1786 reconciliation stalled through May 22
What’s Coming — added “Gemini Omni Pro shipping window”: New item. Google announced Pro at I/O with no shipping date; “soon” with the timer running. Independent benchmarks for Omni Flash against Veo 3.1, Kling 3.0, and HappyHorse-1.0 are the test that converts the launch keynote’s world-model framing from rhetoric to evidence — or not
Considered and excluded: Agentic-orchestration layer Stack section (deferred per operator instruction to the 2026-05-26 ops meeting — existing standing-watch item governs); Kling 2027 IPO target with Tencent anchor (Cowork comparison drop cited a Pandaily URL that 404s; “2027” attributed to internal Kuaishou board documents with no public primary surfacing in WebSearch — hold until a primary lands); Higgsfield as a base-model entry (still an orchestrator, not a base model — same exclusion as 2026-05-18, pending the 5/26 structural decision); Veo brand retirement claim (Cowork-comparison framing; Pichai keynote primary contains no Veo discontinuation language — verified against blog.google/innovation-and-ai/sundar-pichai-io-2026/ during R#5 sourcing pass)
params.toml [landscape]: Bumped commercial = 7 → commercial = 8 to match the new Big Eight count; open_source = 9 unchanged
Paired update outside this file: Synced up_next.date in content/posts/ai-video-weekly-roundup-2026-05-25.md from 2026-05-18 to 2026-05-24 so the Stack’s lastmod precedes the Roundup’s publish date; ordering preserves the Roundup as tomorrow’s lead article
Process note: Phase-2 executed as uninterrupted Edit sequence (same precedent as 2026-05-18) rather than a hand-retyped full-file Write — transcription-risk avoidance on a 499-line living-reference page

May 18, 2026

Last updated date: Advanced from May 10 to May 18, 2026
Tags: Added higgsfield — Higgsfield Supercomputer surfaces as an agentic-orchestration entrant
Kling 3.0 / 3.0 Omni — Kuaishou: Added paragraph on Kuaishou’s May 13 confirmation it is weighing external financing for Kling AI (company-stated as preliminary, no definitive agreements); reported ~$20B target valuation / ~$2B pre-IPO round / Tencent among potential investors / ~$500M annualized revenue flagged explicitly as press attribution, not company-confirmed. Framed as the clearest commercial-scale signal in AI video to date and first pure-play comparable since Sora
Runway Gen-4 Turbo — Runway: Added Runway Agent (May 13, conversational end-to-end production) and the $40M Japan expansion (Tokyo office; enterprise base tripled — Yamaha/SoftBank/NHN); updated Best for tagline, Key features, Access, and Next bullets; positioned within the agentic-orchestration layer
HappyHorse-1.0: Updated open-weights framing from “two weeks” to week-three current status (no movement, no ship date, marketing claim unchanged) in both the model entry and the What’s Coming item
How to Choose routing framework: Broadened the multi-model-agent route to include Higgsfield Supercomputer alongside Pika Agents; noted Runway Agent for the single-vendor end-to-end case
What’s Coming: Added “Agentic-orchestration layer” item (Luma/Pika/Runway/Higgsfield convergence, with the open structural question of whether it needs its own section); rewrote “TAKE IT DOWN Act platform compliance” to current state (May 19 imminent; FTC’s 15-platform warning letters May 13; $53,088/violation; hash-dedup FTC-recommended-not-mandated; generators unnamed but covered at output stage); added “State synthetic-media provenance mandates” item (CT SB 5 awaiting signature, AZ SB 1786 Senate-passed, HI HB 2137) as a track distinct from the NCII-liability item
Considered and excluded: Apple/HeyGen AI-presenters (single-source, unverified — same exclusion as the roundup); Google I/O “What’s Coming” item left as-is (accurate pre-keynote; Wednesday pulse + next roundup carry the actual reveal); Higgsfield not added as a Big-N model entry (it is an orchestrator, not a base model — tracked as a What’s Coming category pending a structural decision); Quick Reference table and params.toml [landscape] counts unchanged (no model spec/pricing/access change; no model added or removed)
Process note: Phase-2 executed as an uninterrupted Edit sequence (single reviewable git diff, no operator pause) rather than a hand-retyped full-file Write as the skill’s literal text specifies — chosen to eliminate transcription risk on a 477-line living-reference page. Same single-diff review outcome. Flagged for skill update: the landscape-skill should codify “atomic uninterrupted Edit sequence in one Phase-2 turn” as equivalent to the mandated Write

May 10, 2026

Last updated date: Advanced from May 5 to May 10, 2026 (the May 5 lastmod was a Stack rename housekeeping touch; substantive content last edited May 4)
Veo 3.1 — Google DeepMind: Added paragraph on Veo’s April 29 arrival on Gemini-enabled TCL Google TVs in the US — voice-driven generation through Gemini’s Create tab, either from scratch or by animating still images; Nano Banana ships alongside in the same Gemini-on-TV bundle; TCL-only and US-only at launch. Updated Access spec bullet to include the new TV distribution surface alongside Flow / Vids / Adobe Firefly. Updated Key features bullet to mention voice-driven generation on Google TV. Catch-up from the May 5 housekeeping window
What’s Coming: Replaced standalone “Veo 4 expected at Google I/O 2026” item with a unified “Google I/O 2026 (May 19–20)” entry covering both the Veo 4 expectation and the early-May “Omni” leak. A UI string in Gemini’s video tab reading “Powered by Omni” (captured May 2 by TestingCatalog) replaced the prior “Veo 3.1” attribution on the same surface. Three plausible readings — Veo rebrand, parallel new Gemini-trained video model, or unified image-video omni-model collapsing the Veo / Nano Banana split — resolve on stage May 19
What’s Coming: Rewrote TAKE IT DOWN Act item with full compliance bar (48-hour removal window from valid notice, FTC Section 5 enforcement, Section 230 non-shield, nonprofits in scope) and the observation that no major AI video lab — Runway, Kling, Veo, Pika, HappyHorse, LTX — has published a compliance page as of May 8. Deadline is nine days out at update time
What’s Coming: Rewrote HappyHorse-1.0 open-source weights item with explicit “no movement in two weeks” framing — empty public GitHub (no weights, no inference code, no license), “coming soon” Hugging Face, and an ATH marketing claim that hasn’t softened or acknowledged the gap. If weights don’t ship before Google I/O, expect “coming soon” through summer to be the operational posture
What’s Coming — added (same-day re-touch after Cowork comparison): New “White House FDA-style AI vetting EO” item — NEC Director Hassett confirmed May 6 (Bloomberg, CNBC) the White House is drafting an EO requiring frontier AI models to pass federal pre-release review on an FDA-drug-approval model; Google, Microsoft, xAI named in-scope; “next two weeks” window; Anthropic’s Mythos model the policy trigger. Would reset launch calculus for every commercial video model at frontier scale
What’s Coming — added (same-day re-touch after Cowork comparison): New “State deepfake legislation map” item — Connecticut HB 5312 (private right of action against creators of nonconsensual AI-generated intimate imagery), Vermont election-deepfake bill, Iowa chatbot-safety law, Utah’s nine AI bills to Cox on May 6, joining Tennessee ELVIS Act, California AB 2655, and New York’s election-deepfake law. Federal preemption isn’t happening this year; AI video labs need to model state civil exposure alongside federal compliance

May 4, 2026

Last updated date: Advanced from April 27 to May 4, 2026
Tags: Added fal, minimax, pika-agents — fal is now a primary distribution surface for HappyHorse-1.0; MiniMax surfaces as one of the orchestrated models inside Pika Agents; Pika Agents is its own product line worth tagging separately from pika
Quick Reference table: Rewrote HappyHorse-1.0 row to reflect commercial API launch (April 27) — Max Resolution 1080p (joint audio, 7-language lip-sync), Free Tier Weights pending; API live, Paid From $0.14/sec 720p · $0.28/sec 1080p, API ✓ via fal.ai + Alibaba Cloud Bailian. Renamed Pika row to “Pika 2.5 / Pika Agents” with Best For updated to flag the multi-model orchestration layer. Updated Grok Imagine row resolution annotation to (Pro 1080p delayed past April)
HappyHorse-1.0 — promoted to standalone entry: Moved from a single bullet in “Other Notable Open-Source Models” to a full standalone section within Open-Source & Local Generation. Covers the April 7 anonymous reveal, April 10 Alibaba unmasking, April 27 commercial API launch on fal (4 endpoints, $0.14/sec 720p, $0.28/sec 1080p) plus Alibaba Cloud Bailian enterprise rollout, current Elo 1,354 T2V / 1,392 I2V (no-audio leaderboards), Elo 1,218 audio-included T2V (#2, near-tied with Seedance at 1,220), 15B-parameter unified Transformer architecture, joint audio-video in a single forward pass, seven-language native lip-sync, ~38s inference for 1080p on a single H100. Open-weights credibility section: ATH’s “fully open-sourced” landing-page claim diverges from the empty GitHub repo; treat HappyHorse-1.0 as a commercial model with an open-source promise until weights actually ship
Pika 2.5 — Pika Labs: Added Pika Agents (April 28) paragraph — multi-modal AI creative partner orchestrating Pika Video, Seedance 2.0, Kling, MiniMax, Veo 3, and Sora; audio via ElevenLabs / MiniMax / Whisper; 17+ platform surfaces (Slack, Telegram, Discord, X, Notion, Figma, Zoom, etc.). Closes the “Pika blog quiet” thread flagged in the April 27 update — Pika answered with the agent product. Renamed entry heading to “Pika 2.5 / Pika Agents.” Updated Best For tagline, Audio spec bullet, Key feature bullet, and Access bullet to reflect the agent product alongside Pika 2.5 and PikaStream 1.0
Grok Imagine — xAI: Added paragraph on the 1080p Pro tier missing its late-April commitment and slipping into May with no new public timeline. Reframed the competitive context: Veo 3.1 Lite ships 1080p at $0.08/sec, Kling 3.0 ships native 4K at $8/mo, Pika 2.5 ships 1080p at $8/mo, even open-source LTX-2.3 outputs true 4K — every week Pro slips, the per-minute price advantage matters less and the resolution gap matters more. Updated Max resolution spec, Next bullet, and linked the May 4 roundup
Other Notable Open-Source Models: Replaced HappyHorse-1.0 paragraph with a brief “promoted above” pointer. Pulled Happy Oyster out into its own bullet, placed adjacent to HunyuanVideo / HY-World 2.0 since both are 3D world models from Chinese labs released April 16
How to Choose routing framework: Added “Need top benchmark quality with commercial API access?” → HappyHorse-1.0 via fal.ai or Alibaba Cloud Bailian. Added “Need a multi-model creative agent that orchestrates Kling, Veo, Seedance, MiniMax, and Sora from a conversational interface?” → Pika Agents
What’s Coming: Added “Veo 4 expected at Google I/O 2026” item (May 19–20; Google has used I/O for Veo announcements in both 2024 and 2025). Updated “Grok Imagine Pro (1080p)” item to reflect the late-April slip. Updated “HappyHorse-1.0 open-source weights” item with the credibility-test framing — ATH’s “fully open-sourced” marketing claim diverges from the empty GitHub repo state; commercial API launched separately

April 27, 2026

Last updated date: Advanced from April 17 to April 27, 2026
Tags: Added adobe, firefly, hunyuan, world-models — Adobe Firefly’s multi-model hub is now a primary distribution surface for tracked models; Tencent HY-World 2.0 establishes Hunyuan as a separate open-source thread; world-models is now a recurring beat
Quick Reference table: Updated Veo 3.1, Kling 3.0, and Runway Gen-4 Turbo rows to flag Adobe Firefly as an additional access channel; renamed Kling row to “Kling 3.0 / 3.0 Omni” to surface the new variant; renamed Runway row to “Runway Gen-4 Turbo / Gen-4.5” to reflect Gen-4.5’s appearance via Firefly
Sora 2 — OpenAI: Updated status line and shutdown timeline from future-tense to past-tense — consumer app and web interface went dark on April 26, 2026, as scheduled. API access still continues through September 24, 2026. Linked to April 27 weekly roundup
Luma Ray 3.14: Added Boundless / Mazda commercial detail — Johannesburg agency delivered Mazda’s first AI-produced commercial in under two weeks using Luma Agents (April 2026); positioned as the most credible production-deployment signal for any AI video platform this year
Kling 3.0 → Kling 3.0 / Kling 3.0 Omni: Renamed entry; added Kling 3.0 Omni variant with shot/camera/character controls; added paragraph on April 15 Adobe Firefly hub addition; updated Access spec to include Adobe Firefly distribution
Veo 3.1 — Google DeepMind: Updated Access spec to note Adobe Firefly multi-model hub availability (April 15)
Runway Gen-4 Turbo: Updated description to note Gen-4.5 is now accessible via Adobe Firefly (April 15) — Runway’s first major distribution beyond its own platform; updated Access, Next bullets, and What’s Coming entry accordingly
Grok Imagine — xAI: Added paragraph on Grok 4.3 Beta native video understanding (April 17); paired generation + understanding under one model family; updated Key feature bullet
Pika 2.5 — Pika Labs: Added PikaStream 1.0 (April 2) — real-time AI video engine for live agent meetings (24fps/480p, ~1.5s latency); flagged Pika blog quiet period since launch; updated spec bullets to reflect both Pika 2.5 and PikaStream
Other Notable Open-Source Models: Renamed HunyuanVideo entry to “HunyuanVideo / HY-World 2.0” — added Tencent’s April 16 multi-modal world model release with editable 3D scenes (meshes + Gaussian Splattings) and open weights; positioned as the more pipeline-friendly counterpart to Alibaba’s gated Happy Oyster
How to Choose routing framework: Added “Need real-time live AI video for agent meetings?” → PikaStream 1.0; added “Need Adobe Creative Cloud integration?” → Firefly multi-model hub; added “Need editable 3D world / scene generation?” → HY-World 2.0 or Happy Oyster; expanded multi-shot routing to include Kling 3.0 Omni
What’s Coming: Updated Runway Gen-4.5 entry — now in market via Firefly, standalone runwayml.com launch still pending; updated OpenAI / Sora item to reflect April 26 app shutdown executed and remaining September 24 API sunset; added “Adobe Firefly multi-model expansion” item; added “Tencent vs. Alibaba 3D world model race” item

April 20, 2026

HappyHorse-1.0: T2V (no audio) Elo updated to 1,364 (from 1,361 as of April 17)

April 17, 2026

Last updated date: Advanced from April 11 to April 17, 2026
Tags: Added skyreels tag to reflect SkyReels V4 promotion to tracked model
Quick Reference table: Updated Wan 2.2 → Wan 2.7 with new capabilities (Thinking Mode, 5 task types, ComfyUI/Model Studio access); added SkyReels V4 row (1080p, 70 free credits/mo, open source); updated Grok Imagine resolution to note “1080p Pro coming”
Wan 2.2 → Wan 2.7: Complete rewrite of model entry. Wan 2.7 released April 3 with Thinking Mode (chain-of-thought reasoning), 5 unified task types (T2V, I2V, video continuation, reference-to-video, video editing), ComfyUI 0.18.5 support, 1080p at 2–15 seconds
SkyReels V1 → SkyReels V4: Updated from V1 to V4 in Other Notable Open-Source Models. SkyReels V4 (April 3) is the first open-source model with joint audio-video generation via dual-stream MMDiT; 1080p/32FPS, up to 15 seconds; Elo ~1,135 on Artificial Analysis T2V with audio; 70 free monthly credits
HappyHorse-1.0: Updated Elo scores (T2V no audio: 1,361, up from 1,333; I2V no audio: 1,398, up from 1,392); added 15B parameter count and 40-layer unified Transformer architecture detail; added Happy Oyster world model release (April 16) — ATH’s interactive 3D environment generator for gaming, film, and VR
Grok Imagine: Updated resolution spec to note 1080p Pro confirmed for late April 2026; added Quality + Speed dual generation modes (early April); updated engine description to “Aurora autoregressive MoE model”; added Next bullet for Pro tier; updated Access to include SuperGrok
What’s Coming: Added “Grok Imagine Pro (1080p)” item — confirmed by Musk for late April, available to SuperGrok subscribers

April 11, 2026

Last updated date: Advanced from April 4 to April 11, 2026
Seedance 2.0 Pro: Updated benchmark position — no longer top-ranked on Artificial Analysis; displaced to #2 T2V by HappyHorse-1.0 (Elo 1,333) as of April 10. Updated US access status: Dreamina Seedance 2.0 now rolling out in US CapCut with real-face image-to-video disabled, unauthorized IP blocked, and invisible watermarking on all output. Updated Access, API, and Note spec bullets accordingly. Updated description paragraph to reflect that US was excluded but is now partially available
Veo 3.1 — Google DeepMind: Confirmed Veo 3.1 Fast price reduction occurred April 7 as committed; removed “scheduled for April 7” language, updated API bullet to direct developers to Gemini API docs for current per-second rates; updated description paragraph to reflect the cut landed
Other Notable Open-Source Models: Added HappyHorse-1.0 (Alibaba ATH AI Innovation Unit) — #1 T2V at Elo 1,333, #1 I2V at Elo 1,392 on Artificial Analysis; model in beta, weights and GitHub “coming soon”; expected to become the top open-source video model on release
How to Choose routing framework: Updated Seedance routing entries to remove “(if accessible)” and add “via CapCut (US, with restrictions)” to reflect partial US access
What’s Coming: Removed “Seedance 2.0 Pro US access” (resolved partially via CapCut with restrictions); removed “Veo 3.1 Fast price cut April 7” (confirmed delivered); added “HappyHorse-1.0 open-source weights” (coming soon per ATH team); added “TAKE IT DOWN Act platform compliance — May 19, 2026”; updated “Seedance copyright litigation” item to reflect current state with US CapCut access but unresolved studio disputes

April 4, 2026

Last updated date: Advanced from March 27 to April 4, 2026
Sora 2 — OpenAI (discontinued): Added specific shutdown timeline — app and web interface off April 26, 2026; API access continues until September 24, 2026
Veo 3.1 — Google DeepMind: Major access and pricing update. Added Google Vids free tier (10 clips/month for any Google account, 8 sec at 720p, text-to-video and image-to-video); added Google AI Pro/Ultra features (1,000 clips/month, Lyria 3 music up to 3 min, AI avatars with scene/wardrobe control, YouTube export now live). Added Veo 3.1 Lite developer tier via Gemini API / Google AI Studio ($0.05/sec 720p, $0.08/sec 1080p, <50% cost of Fast tier, same speed). Updated “Best for” tagline, description paragraphs, Access spec, and API spec bullets. Updated Key features bullet to reflect Vids integration
How to Choose routing framework: Added “Need a free starting point? → Veo 3.1 via Google Vids (10 free clips/month)”
What’s Coming: Removed “Google Flow + YouTube integration” (YouTube export now live in Google Vids); added “Veo 3.1 Fast price cut April 7”; updated “OpenAI robotics/world simulation” item to include Sora API September 24 deadline; replaced with “Google Vids/Workspace expansion” item for ongoing paid tier rollout

March 27, 2026

Last updated date: Advanced from March 20 to March 27, 2026
Tags: Added luma tag to reflect Luma Ray 3.14 promotion to primary tracked model
Sora 2 — OpenAI: Removed from active Big Seven tracking — product discontinued March 24, 2026. Section replaced with a brief discontinuation note citing $15M/day burn rate vs. $2.1M lifetime revenue, Disney deal collapse, and OpenAI’s pivot to robotics/world simulation. Model entry retained as a historical reference with strikethrough heading
Luma Ray 3.14 — Luma AI: Added as new Big Seven entry, replacing Sora. Native 1080p, 4× faster generation, 3× cheaper per-second pricing vs. previous Ray. Ray3 Modify companion tool for hybrid performance/acting workflows. $900M Series C (HUMAIN), new London office. Enterprise deployments via Luma Agents (Publicis, Adidas, Mazda). Promoted from “Models to Watch” list
Seedance 2.0 Pro: Added paragraph on Dreamina Seedance 2.0 relaunch (March 23) in Africa, South America, Middle East, Southeast Asia — US still excluded. Added paragraph on three-way Washington copyright standoff: Blackburn bill vs. White House AI framework vs. CLEAR Act (Schiff/Curtis). Updated Access and Note spec bullets to reflect Dreamina regional launch
How to Choose routing framework: Removed “Need realistic physics? → Sora 2”; added “Need professional production volume at scale? → Luma Ray 3.14”; updated “character consistency” and “multi-shot narrative” routes to include Luma Ray 3.14 alongside Seedance as a US-accessible alternative
What’s Coming: Removed “Sora integration into ChatGPT” (product dead); added “White House AI framework vs. CLEAR Act” (new copyright legislative developments); updated Seedance item to reflect Dreamina emerging-markets launch and US exclusion; added “OpenAI robotics/world simulation” item tracking what comes next from OpenAI in the video-adjacent space

March 20, 2026

Last updated date: Advanced from March 13 to March 20, 2026
Tags: Added regulation tag — regulatory developments are now a recurring weekly topic affecting multiple tracked models
Runway Gen-4 Turbo: Added third description paragraph covering the GTC real-time video model research preview (sub-100ms TTF on Vera Rubin hardware, March 17) and Gen-4.5 signal (confirmed running on Vera Rubin ahead of public release, no launch date); added Next: spec bullet summarizing both developments
Seedance 2.0 Pro: Added paragraph noting legislative reinforcement from the Blackburn draft AI bill (March 19) — which, if enacted, would make unauthorized AI training use of copyrighted works categorically not fair use; updated Note: spec bullet to reference the bill
What’s Coming: Removed “NVIDIA GTC 2026” item (event concluded); added “Runway Gen-4.5” (previewed at GTC); added “NVIDIA Vera Rubin cloud deployment H2 2026” (confirmed by AWS, Google Cloud, Microsoft, OCI; 10× inference cost reduction); added “DLSS 5 Fall 2026” (neural rendering with VFX/filmmaking ambitions beyond gaming); added “Blackburn draft AI bill” (copyright/deepfake provisions); updated Seedance item to note Blackburn bill adds legislative pressure

March 13, 2026

Last updated date: Advanced from March 7 to March 13, 2026
Sora 2 (OpenAI): Added Sora 1 retirement notice (effective March 13, US users); added note on reported ChatGPT integration plans (The Information, March 12) and declining standalone app install figures (−45% MoM, January 2026); updated Disney licensing detail to specify all four brands (Disney, Marvel, Pixar, Star Wars); added Note spec bullet summarizing access changes
Runway Gen-4 Turbo: Added $315M Series C / $5.3B valuation (February 2026); added two-paragraph description of Runway Characters (launched March 9) — real-time avatar API built on GWM-1, BBC and Silverside as launch partners, available at dev.runwayml.com; updated Best for tagline and spec bullets to reflect Characters alongside Gen-4 Turbo
LTX-2 → LTX-2.3: Renamed entry to reflect new version; rewrote description to cover the 22B-parameter rebuild (new VAE, 4× text connector, HiFi-GAN vocoder), desktop video editor launch, portrait mode (9:16), last-frame interpolation, and 24/48fps options; updated NVIDIA acceleration figures from “3× / 60%” (CES) to “2.5× / 60%” (GDC RTX 50 Series specific); added RTX Video Super Resolution 30× upscaling stat; added ComfyUI App View simplified UI; updated spec bullets (true 4K native, 50fps, Apache 2.0 commercial licensing threshold)
Other Notable Open-Source Models: Added Helios (Peking University / ByteDance / Canva) — 14B autoregressive diffusion model, 19.5fps real-time on single H100, Apache 2.0, released March 2026
How to Choose routing framework: Updated “LTX-2” reference to “LTX-2.3”; added “Need real-time interactive avatars? → Runway Characters (GWM-1)”
What’s Coming: Added “Sora integration into ChatGPT” item; updated “NVIDIA GTC 2026” from “later this month” to specific March 16–19 date with Jensen Huang keynote detail; all other items unchanged