Models / Video
MiniMax H3
By MiniMax, in Claude, ChatGPT, Cursor, Claude Code, and the API.
The lowest-cost video, and the best in blind tests at animating a reference image. 768p, without sound.
- Made by
- MiniMax
- Type
- Video
- Price band
- Budget
- Blind tests
- #3 on Artificial Analysis's text-to-video arena, September 2026
- Lengths
- 5–15 seconds, default 5
- Sound
- No
- Shapes
- landscape, portrait
- Reference image
- The clip starts from your image
- Model id
- minimax-h3
Ask for it by name
Your AI uses the default for video unless you name another. Say MiniMax H3 and it uses this one.
youMake a 5-second MiniMax H3 clip of waves breaking at dusk.
A Studio month (3,000 credits) covers about 50 5-second clips with MiniMax H3, or a mix with any other model. See plans.
From your own code
The same key your assistant uses works as a bearer token. See the docs.
curl https://getstudiopass.com/v1/generations \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"modality":"video","model":"minimax-h3","prompt":"Waves breaking at dusk","duration_seconds":5}'