MiniMax has open-sourced its new flagship AI model, MiniMax-M2, positioning it as one of the most efficient coding and agentic AI systems currently available.
Built as an “Agent & Code Native” model, MiniMax-M2 is designed for end-to-end developer workflows and agentic reasoning.
Despite having 230 billion total parameters, it activates only 10 billion at a time, enabling near frontier-level performance in a more compact, cost-efficient form.
MiniMax claims that the model delivers results at roughly 8 per cent of the cost of Claude Sonnet and runs nearly twice as fast. According to the Artificial Analysis Intelligence Index v3.0, MiniMax-M2 achieved a score of 61, ranking eighth overall and outperforming Anthropic’s Claude Opus 4.1, which scored 59.
The Artificial Analysis benchmark combines results from 10 key evaluations including MMLU-Pro, GPQA Diamond, AIME 2025, SciCode, and Terminal-Bench Hard, to assess general reasoning and tool-use performance.
MiniMax-M2 stands among the strongest open-source models on the leaderboard, placing above Qwen 3 72B (58) and DeepSeek-V3.2 (57). While not the top open-source model, it ranks among the highest-performing publicly available ones in this benchmark.
Benchmark comparisons show its coding performance to be highly competitive, scoring 46.3 on Terminal-Bench, surpassing Claude Sonnet 4.5 and Gemini 2.5 Pro, and 44 on BrowseComp, well ahead of Claude Sonnet 4.5’s 19.6.
MiniMax has made MiniMax-M2 free to use for a limited time via its Agent and API platforms, and has also open-sourced the model weights on Hugging Face and GitHub for local deployment.
With benchmark results placing it above Claude Opus 4.1, MiniMax-M2 reinforces the growing strength of open-source AI models that aim to balance affordability, speed, and advanced reasoning in real-world coding and agentic applications.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.