MiniMax-M2
MiniMax-M2, a model built for Max coding & agentic workflows.
What is MiniMax-M2?
MiniMax-M2, a model built for Max coding & agentic workflows.
Overview of MiniMax-M2
What it is
MiniMax-M2 is an open-source Mixture-of-Experts (MoE) language model released by MiniMax. It has 230 billion total parameters with 10 billion active parameters per forward pass. The model is designed for coding and agentic workflows, aiming to provide efficient performance in a compact form factor. Its composite intelligence score ranks #1 among open-source models globally according to Artificial Analysis benchmarks.
Key features
- Superior intelligence: competitive general intelligence across mathematics, science, instruction following, coding, and agentic tool use.
- Advanced coding: excels at multi-file edits, coding-run-fix loops, and test-validated repairs, with strong performance on Terminal-Bench and (Multi-)SWE-Bench–style tasks.
- Agent performance: plans and executes complex, long-horizon toolchains across shell, browser, retrieval, and code runners; performs well on BrowseComp-style evaluations.
- Efficient design: 10 billion activated parameters deliver lower latency, lower cost, and higher throughput for interactive agents and batched sampling.
- Open-source: released under the MIT license.
Use cases
MiniMax-M2 is suited for end-to-end developer workflows such as editing real repositories, executing commands, browsing the web, and delivering functional solutions. It can be used in terminals, IDEs, and CI environments across multiple programming languages. The model is also applicable for agentic tasks involving shell, browser, retrieval, and code runner toolchains.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
The model is released under the MIT license. Pricing is not stated in the provided materials.
Based on the project's public README and repository metadata.