MiniMax
MiniMax models you can run locally
MiniMax develops multimodal language models and open video generation systems, including MiniMax M3 and H3.
Available models
Ordered from the lightest model to the largest.
MiniMax H3
33B · Dense
Open video generation — text/image to 2K video with native stereo audio
32K ctx
MiniMax H3 Image
33B · Dense
Open 16-megapixel image generation and editing — 6-step distilled H3 DiT, up to 16 reference images
4K ctx
MiniMax M3
428B · 23B active · MoE
Native multimodal MoE — understands text, image and long video with 1M context
1024K ctx
MiniMax M3.1 Flash
428B · 23B active · MoE
M3 update for agentic coding, computer use and cybersecurity — DSpark speculative decoding, 1M context
1024K ctx