MiniMax
MiniMax M3
Open-weights with video input and a 512K output window.
Open weightstextimagevideoREL 2026-05-31
Compare this model →Verified 2026-06-12
Overview
MiniMax M3 is an open-weights model with an unusual combination: 1M context, text+image+video input, and up to 512K output tokens, priced at $0.30/$1.20 per 1M. A strong pick for long-form generation and video understanding on a budget.
Benchmark scores
Methodology →Intelligence55
Codingn/a
Reasoningn/a
Mathn/a
Sources and verification
- Intelligence: Artificial Analysis Intelligence Index · checked 2026-06-12
- Catalogue and pricing: OpenRouter model record · checked 2026-06-12
Strengths and weaknesses
Strengths
- Video input — rare among open models
- 512K max output
- Open weights, cheap
Trade-offs
- Less ecosystem tooling than DeepSeek/Qwen
- Mid-tier reasoning
Reach for this model when
Long-form generation and budget video-understanding workloads.
Run something like it locally
Self-hostable via vLLM on multi-GPU hardware.
