Picsum ID: 198
MiniMax-M2: The New Leader in Open-Source LLMs, Especially for Agentic AI
In the rapidly evolving world of artificial intelligence, a new contender has emerged, challenging established players like DeepSeek and Qwen. This newcomer is MiniMax-M2, the latest large language model (LLM) from the Chinese startup MiniMax. What sets MiniMax-M2 apart is its impressive performance, particularly in agentic tool use. This capability allows the model to interact with other software, such as web search and custom applications, with minimal human intervention. Moreover, MiniMax-M2 is available under a permissive, enterprise-friendly MIT License, making it freely accessible for developers to use, modify, and deploy—even for commercial purposes.
Key Features and Availability
MiniMax-M2 is readily available on platforms like Hugging Face, GitHub, and ModelScope. It also offers an API through MiniMax, supporting OpenAI and Anthropic API standards. This compatibility simplifies the transition for users of proprietary AI models from other providers.
Performance and Benchmarks
Independent evaluations by Artificial Analysis, a generative AI model benchmarking organization, place MiniMax-M2 at the forefront of open-weight systems. It leads in the Intelligence Index, a composite measure of reasoning, coding, and task execution. In agentic benchmarks, which assess a model’s ability to plan, execute, and use external tools, MiniMax-M2 achieves remarkable scores. These scores put it on par with top proprietary systems like GPT-5 (thinking) and Claude Sonnet 4.5, making it the highest-performing open model for real-world agentic and tool-calling tasks.
What it Means for Enterprises
Built on an efficient Mixture-of-Experts (MoE) architecture, MiniMax-M2 delivers high-end capabilities for agentic and developer workflows while remaining practical for enterprise deployment. This marks a turning point for open models in business settings. It combines advanced reasoning with a manageable activation footprint of just 10 billion active parameters out of 230 billion total. This design enables enterprises to operate advanced reasoning and automation workloads on fewer GPUs, reducing infrastructure demands and licensing costs associated with proprietary systems.
Competitive Advantages
MiniMax-M2 excels in benchmarks for end-to-end coding, reasoning, and agentic tool use, often leading or closely trailing proprietary systems. Its performance in tests like τ²-Bench, SWE-Bench, and BrowseComp highlights its advantages for organizations relying on AI systems for complex workflow planning, execution, and verification—essential functions for agentic and developer tools.
Technical Architecture and Scalability
MiniMax-M2’s technical architecture is a sparse Mixture-of-Experts model with 230 billion total parameters and 10 billion active per inference. This configuration reduces latency and compute requirements while maintaining broad general intelligence. The design facilitates responsive agent loops that execute faster and more predictably than denser models. For enterprise technology teams, this means easier scaling, lower cloud costs, and reduced deployment friction. According to Artificial Analysis, the model can be served efficiently on as few as four NVIDIA H100 GPUs at FP8 precision.
Benchmark Leadership
MiniMax-M2 demonstrates strong real-world performance in developer and agent environments. It achieves top or near-top performance in various categories, including SWE-bench Verified, ArtifactsBench, τ²-Bench, GAIA (text only), BrowseComp, and FinSearchComp-global, showcasing its capability in executing complex, tool-augmented tasks across multiple languages and environments.
Overall Intelligence Profile
The model’s overall intelligence is confirmed in the latest Artificial Analysis Intelligence Index v3.0, where it scored 61 points, ranking as the highest open-weight model globally, closely following GPT-5 (high) and Grok 4. This consistency indicates a reliable foundation suitable for software engineering, customer support, or knowledge automation systems.
Designed for Developers and Agentic Systems
MiniMax engineered M2 for end-to-end developer workflows, enabling multi-file code edits, automated testing, and regression repair. It also excels in agentic planning, handling tasks that combine web search, command execution, and API calls while maintaining reasoning traceability. These capabilities make MiniMax-M2 especially valuable for enterprises exploring autonomous developer agents, data analysis assistants, or AI-augmented operational tools. Benchmarks such as Terminal-Bench and BrowseComp demonstrate the model’s ability to adapt to incomplete data and recover gracefully from intermediate errors, improving reliability in production settings.
Interleaved Thinking and Structured Tool Use
A distinctive aspect of MiniMax-M2 is its interleaved thinking format, maintaining visible reasoning traces between tags. This enables the model to plan and verify steps across multiple exchanges, a critical feature for agentic reasoning. The company provides a Tool Calling Guide on Hugging Face, detailing how developers can connect external tools and APIs via structured XML-style calls. This functionality allows MiniMax-M2 to serve as the reasoning core for larger agent frameworks, executing dynamic tasks such as search, retrieval, and computation through external functions.
Open Source Access and Deployment Options
Enterprises can access the model through the MiniMax Open Platform API and MiniMax Agent interface, both currently free for a limited time. MiniMax recommends SGLang and vLLM for efficient serving, each offering day-one support for the model’s unique interleaved reasoning and tool-calling structure. Deployment guides and parameter configurations are available through MiniMax’s documentation.
Cost Efficiency and Token Economics
MiniMax’s API pricing is competitive, set at $0.30 per million input tokens and $1.20 per million output tokens. This cost-performance balance is an advantage for teams deploying interactive agents or high-volume automation systems.
Background on MiniMax
MiniMax has quickly risen in China’s AI sector, backed by Alibaba and Tencent. The company gained international recognition through breakthroughs in AI video generation and open-weight LLMs designed for developers and enterprises. The company first captured global attention in late 2024 with its AI video generation tool, “video-01,”. CEO Yan Junjie emphasized that the system outperformed leading Western tools in generating human movement and expression. In early 2025, MiniMax unveiled the MiniMax-01 series, including MiniMax-Text-01 and MiniMax-VL-01, introducing an unprecedented 4-million-token context window. The company continued its rapid cadence with the MiniMax-M1 release in June 2025.
Open-Weight Leadership and Industry Context
The release of MiniMax-M2 reinforces the growing leadership of Chinese AI research groups in open-weight model development. Artificial Analysis observed that MiniMax-M2 exemplifies a broader shift in focus toward agentic capability and reinforcement-learning refinement, prioritizing controllable reasoning and real utility over raw model size. MiniMax-M2 is positioned as a practical foundation for intelligent systems that think, act, and assist with traceable logic—making it one of the most enterprise-ready open AI models available today.
About the Author
