MiniMax-M2: New Open-Source AI Tops Claude Opus 4.1
MiniMax-M2 Arrives: New Open-Source AI Outperforms Claude Opus 4.1 on Key Benchmarks
In a significant move for the open-source community, AI company MiniMax has released its new flagship model, MiniMax-M2.
Designed from the ground up for elite coding and agentic workflows, the model is already making waves by outperforming established competitors like Anthropic’s Claude Opus 4.1 in critical intelligence indexes.
MiniMax is positioning its new model as an “Agent & Code Native” system, engineered to handle complex, end-to-end developer tasks and sophisticated reasoning.
The model’s architecture introduces a new standard for efficiency: while it contains 230 billion total parameters, it only activates 10 billion at any given time.
This Mixture-of-Experts (MoE) design allows it to deliver near frontier-level performance in a compact, fast, and cost-effective package.
According to the company, MiniMax-M2 operates at roughly 8% of the cost of Claude Sonnet and runs nearly twice as fast, a compelling proposition for developers and businesses alike.
This efficiency does not come at the cost of power. In the latest Artificial Analysis Intelligence Index v3.0-a comprehensive benchmark that evaluates models on reasoning, tool use, and coding-MiniMax-M2 achieved a score of 61.
This places it eighth overall and notably ahead of Claude Opus 4.1, which scored 59, cementing its status as a powerful new contender in the AI landscape.
A New Leader in Coding Performance
When measured on benchmarks that simulate real-world developer environments, MiniMax-M2 demonstrates highly competitive capabilities.
On Terminal-Bench, which tests a model’s ability to execute tasks in a terminal environment, it scored 46.3, surpassing both Claude Sonnet 4.5 and Gemini 2.5 Pro.
Its performance in web navigation and information retrieval is equally impressive, scoring 44 on the BrowseComp benchmark-a figure that starkly outperforms Claude Sonnet 4.5’s score of 19.6.
These results highlight the model’s practical effectiveness for the daily tasks faced by programmers and automated systems.
Commitment to Open-Source Accessibility
In line with its goal of empowering the broader AI community, MiniMax has made the model widely accessible. For a limited time, MiniMax-M2 is available for free use via its Agent and API platforms.
Furthermore, the company has open-sourced the model weights on both Hugging Face and GitHub, allowing researchers and developers to deploy and fine-tune the model locally.
By delivering a model that excels in performance while championing affordability and speed, the release of MiniMax-M2 reinforces the growing strength of open-source AI.
It stands as a powerful example of the innovation happening outside of closed, proprietary systems, offering the world a tool that is both powerful and accessible.
