Open-weight video is now 3.3 Elo behind the best closed model
Artificial Analysis places MiniMax H3 at #2 on its video board with 1241.5, some 3.3 Elo behind Gemini Omni Flash. The gap between the leading open-weight video model and the leading closed one is now inside the margin most buyers could detect.
Three point three Elo is not a capability gap. It is a rounding error that happens to sit on a leaderboard, and it means the open/closed distinction in video generation has stopped describing quality and started describing licensing.
The specification is unusually complete for an open release: 4 to 15 second outputs at 24fps with 32kHz stereo audio generated alongside the frames, up to 2K through a dedicated regeneration path, plus motion transfer, reference-driven generation and generative video editing. Full-modal input — text, image, video and audio understood in one context — rather than a text-to-video pipe with extras bolted on.
Text-to-image took roughly two years to go from closed-only to open-parity. Video has done it faster, and the reason is that the video labs published their architectural choices while competing on training data and compute — the same dynamic that closed the gap in language models.
What that leaves closed video labs selling is reliability, moderation, indemnity and integration. Those are real products. They are just not the product anyone thought they would be selling.
AIbase — Visual Large Models Receive a Major Open-Source Announcement → · WaveSpeed — AI Video Generation News: 2026 Latest Models & Updates →