MiniMax-M2 is an open-sourceMoE model marketed for coding and agentic workflows, with 230B total parameters and 10B active parameters. It provides local deployment options via HuggingFace and multiple serving guides, plus public product and API access for limited-time free usage.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
MiniMax-M2 is described as a compact, fast MoE model built for coding and agentic workflows, targeting end-to-end developer tasks and tool use. It specifies 230 billion total parameters with 10 billion active parameters.
How it works
The repository emphasizes an activation-size strategy (10B activated parameters) to enable a plan → act → verify loop for agentic workflows, aiming for lower latency and higher throughput. It provides benchmarks and comparisons across various tasks and tool use scenarios.
Getting started
- Local deployment paths are outlined with multiple serving frameworks:
- SGLang: Serve MiniMax-M2 using SGLang (SGLang Deployment Guide linked).
- vLLM: Serve MiniMax-M2 with vLLM (deployment guides linked).
- MLX-LM: Serve MiniMax-M2 with MLX-LM (deployment guide linked).
- Inference parameters recommended:
temperature=1.0,top_p = 0.95,top_k = 40. - Model weights are available on HuggingFace: https://huggingface.co/MiniMaxAI/MiniMax-M2.
Recent releases
- RELEASES (latest 0): - none
Traction
- GitHub stats: stars_7d not provided in FACTS; stars: 2601, forks: 216, open_issues: 54. No daily trend data is present, so only raw numbers are listed here.
Behind the repo
- Community and ecosystem links are provided (WeChat, Discord, MiniMax Agent, API docs, MCP, website, HuggingFace, ModelScope). No founder or company-specific details are given beyond project ecosystem references.
Caveats
- License: MIT is shown as a link in the README image; the FACTS field says license: none listed. The repository page lists a license image with MIT, but the structured facts here do not explicitly confirm a license field.
- Created: 2025-10-25; last_push: 2025-11-13. No explicit security advisories or open issues beyond open_issues: 54 are listed in FACTS.