MiniMax-M1
Open-source reasoning LLM for self-hosted use through vLLM or Transformers, with million-token context and function calling under Apache 2.0.
MiniMax-M1 is an open-source reasoning model for developers building agents or working on complex software and mathematical problems. Its million-token context window makes it a candidate for tasks with long inputs that also need extended reasoning. You can serve the model on your own infrastructure through vLLM or use it through Transformers.
The model builds on MiniMax-Text-01 and combines a mixture-of-experts architecture with lightning attention. That attention design reduces computation during long generations compared with the original DeepSeek-R1 in the reported comparison. This matters for workloads where the model spends substantial time reasoning before it produces an answer.
M1 supports function calling: it can identify when a task needs an external function and produce structured arguments for that call. Developers can use this capability as part of an AI agent that interacts with tools. Its reinforcement-learning training covers mathematical problems and software engineering tasks in sandbox environments; general uses include summarization, translation, question answering and creative writing.
The weights use the Apache 2.0 license. Hugging Face hosts the MiniMax-M1-40k and MiniMax-M1-80k variants, which differ in their thinking budgets. vLLM is the recommended backend for production serving, while Transformers provides another deployment path.
