MiniMax H3: The first open-weight AI model to reach #1 worldwide for video content!

MiniMax H3 is a new open-weight model (the training data is public, so anyone can download and run it) from the Chinese company MiniMax — 33 billion parameters, released August 3. In the Artificial Analysis rankings it took 1st place worldwide for AI video editing, and 2nd for generating video straight from text. It makes 4-15 second clips with stereo sound, and you can feed it up to nine reference images in a single request. You can run it on your own hardware instead of going through someone's paid service, and commercial use is allowed for companies making under $20M in revenue.

Until now, the top spots in AI-video rankings belonged only to closed services — Sora, Runway, Kling. This is the first time an open-weight model has broken into 1st place, specifically in the video-editing category. What actually changes for you: because the weights are open, you can fine-tune it (retrain it a bit on your own footage) without paying for anyone's API — that's the real shift, moving control over AI video generation away from big platforms and toward whoever owns the data. I'd read this as a genuine jump for that one slice of video work — editing — not as open models suddenly beating closed ones everywhere; it's still 2nd place, not 1st, in plain text-to-video generation.