Updated 2026-08-15 · Checked daily, updated at least weekly
AI video model release calendar
Every major AI video generation model — released, officially announced, or credibly rumored. We separate the three ruthlessly: if a date isn't confirmed by the developer, it's marked as an estimate with our reasoning. Sources: developer blogs, API provider changelogs (fal.ai, kie.ai), and the Artificial Analysis leaderboard.
Upcoming & rumored
One network for image, video with native audio (up to 20s) and robot action prediction. Video is open to approved applicants only; there is no public API, no published pricing and no open weights yet.
MiniMax says it will publish H3's weights “within the next few days, subject to the relevant laws and regulations”, and designed the model for compatibility with several domestic Chinese accelerators. No repo has appeared on the MiniMaxAI Hugging Face org yet. Would be the strongest openly licensed video model to date.
ByteDance says the Volcano Ark API is coming shortly after the Jul 31 consumer launch. Until it lands there is no per-second price, no third-party endpoint and no access outside Jimeng and Doubao Pro. The three insider-sourced dates before launch (Jul 6, 13, 16) all failed, so treat any new date without an official post the same way.
Two-stage wind-down of Sora 2: the consumer app and sora.com already went offline on Apr 26, 2026; the developer API (sora-2, sora-2-pro) is decommissioned Sep 24, 2026. Teams on Sora endpoints need a migration path before this date.
Not announced. Estimate based on Kuaishou's historical 4–6 month release cadence since Kling 3.0 (Feb 2026).
Confirmed but undated: the open-weight release of the full multimodal backbone — the first openly licensed model generating video, audio and images from one set of weights. No size or license terms announced; FLUX.1/FLUX.2 Dev tiers shipped non-commercial.
Not announced. Google I/O 2026 shipped Gemini-family updates, not Veo 4. Rumors point to native 4K and longer clips.
No Sora 3 has been announced. After shutting down Sora 2, OpenAI said it is repurposing Sora as a world-model research effort — there is no successor product or date on the roadmap.
Shipped
Hailuo's third generation, launched as a general-purpose omni model: one network reads text, images, video and audio as a single context and returns 2K video with native stereo sound, 5–15s. Live the day it was announced — in the Hailuo app, MiniMax's own API and third-party resellers. Open weights promised within days.
Released Jul 31, 2026 after a Jun 23 preview. 30-second single generations with multi-round extension into minutes, up to 30 images + 10 videos + 10 audio as references in one request, and timestamp-addressed editing (green screen, viewpoint, camera move). Live in Jimeng AI and Doubao Pro; the Volcano Ark API is “coming soon” and no 2.5 model ID exists in the ModelArk catalog yet. No international access, no pricing.
Launched Feb 12, 2026 with Doubao 2.0. Multi-modal input (up to 9 images, 3 videos, 3 audio files), native audio, 480p–4K output. ByteDance paused its global rollout in mid-March 2026 after cease-and-desist letters from Disney, Paramount, Warner Bros. and the MPA over copyrighted characters in outputs.
Multi-shot storyboard generation — one prompt produces a sequence of coherent shots.
Google's current flagship video model. Strong physics and native audio; premium API pricing.
How this calendar works
- Released — the model is publicly usable (app or API).
- Announced — the developer has confirmed it officially; date may still be estimated.
- Rumored — unconfirmed. We state our sources and reasoning, never present rumors as fact.
- Retiring — a confirmed shutdown or deprecation with a hard date.