MiniMax M1
MiniMax-M1 is a large-scale, open-weight reasoning model designed for extended context and high-efficiency inference. It leverages a hybrid Mixture-of-Experts (MoE) architecture paired with a custom "lightning attention" mechanism, allowing it to process long sequences—up to 1 million tokens—while maintaining competitive FLOP efficiency. With 456 billion total parameters and 45.9B active per token, this variant is optimized for complex, multi-step reasoning tasks. Trained via a custom reinforcement learning pipeline (CISPO), M1 excels in long-context understanding, software engineering, agentic tool use, and mathematical reasoning. Benchmarks show strong performance across FullStackBench, SWE-bench, MATH, GPQA, and TAU-Bench, often outperforming other open models like DeepSeek R1 and Qwen3-235B.
Specifications
Modalities
Supported Parameters
Max Output Tokens
40,000Data policy
- Prompt retention
- Unknown — we could not confirm
- Training
- Unknown — not declared
"Unknown" does not mean "safe". It means we could not confirm it.
Whether a provider trains on prompts is a declared value from our terms with them. Retention is determined from the upstream listing for every host this model can reach. Neither is guessed.
Frequently asked questions
- Is MiniMax M1 available on FastMetal?
- Not at the moment. MiniMax M3.1 Flash, from the same lab, is available on the FastMetal API today.
- What is the context window of MiniMax M1?
- 1,000,000 tokens, shared between the prompt and the response.
- How does MiniMax M1 rank?
- #218 on the public arena's Japanese board (ELO 1,231). Ranks move as the leaderboard is updated.
Leaderboard
Rankings: LMArena leaderboard dataset (lmarena-ai/leaderboard-dataset), CC BY 4.0, snapshot 2026-10-22; category names mapped and ratings rounded by FastMetal.