MoE coding and agent design
According to the model card, the MoE model totals about 230B parameters with 10B activated and is built for coding and agentic tasks.
Open Source Model Profile · MiniMaxAI
MiniMax-M2
MiniMax-M2 is a 228.69B-parameter MiniMaxM2 text-generation MoE model from MiniMaxAI. According to the model card, about 10B parameters activate per pass for coding and agentic work.
- Publisher
- MiniMaxAI
- Task
- text-generation
- Model type
- minimax_m2
- License
- modified-mit
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
MiniMax-M2 is published by MiniMaxAI as a text-generation model. The captured configuration identifies MiniMaxM2ForCausalLM, and Safetensors metadata reports about 228.69B parameters. According to the model card, it is a compact MoE release with about 230B total parameters and 10B activated for coding and agentic workflows.
Recorded capabilities
Documented evaluation scaffolds
According to the model card, the publisher references Artificial Analysis benchmarks and describes SWE-bench Verified, Multi-SWE-Bench, and SWE-bench Multilingual scaffolds.
Long-horizon tool use
According to the model card, the model plans and executes toolchains across shell, browser, retrieval, and code runners, with BrowseComp-style recovery behavior.
Serving and sampling guidance
According to the model card, SGLang is recommended for serving, with suggested sampling of temperature 1.0, top_p 0.95, and top_k 40.
229B Safetensors record
The captured page text reports 229B parameters with F32, BF16, and F8_E4M3 tensor types and a chat template.
Use cases in the source record
- Coding and agentic workflows using the documented end-to-end tool-use behavior.
- Software-engineering task experiments consistent with the described SWE-bench evaluation scaffolds.
Limitations and unknowns
- No context-window value was extracted from this record; a 128K figure in the card describes an evaluation setup, not a verified model limit.
- The captured license field lists other while the model card text mentions MIT, so the applicable license is unresolved in this evidence.
- Benchmark superiority claims reference Artificial Analysis without extracted scores, so no comparative performance is verified here.
- Provider state is historical snapshot data covering two providers and should be refreshed before being presented as current.
Source and provenance
Source: MiniMaxAI/MiniMax-M2
Captured: Unknown. Processed: 2026-09-07T19:35:41.885370+00:00.
Join Our 💬 WeChat | 🧩 Discord community. MiniMax Agent | ⚡️ API (Now Free for a limited time!) | MCP | MiniMax Website 🤗 Hugging Face | 🐙 GitHub | 🤖️ ModelScope | 📄 License: MIT Meet MiniMax-M2 Today, we release and open source MiniMax-M2, a Mini model built for Max coding & agentic workflows. MiniMax-M2 redefines efficiency for agents. It's a compact, fast, and cost-effective MoE model (230 billion total parameters with 10 billion active parameters) built for elite performance in coding and agentic tasks, all while maintaining powerful general intelligence. With just 10 billion activated parameters, MiniMax-M2 provides the so…
F001F002F003F004F005F006F007F009F010F012F013F014F015F016F017F018F019F020F021F022