MiniMax M2, backed by Alibaba and Tencent, is unlocking access to top-tier AI capabilities with a 230B-parameter open source model.
Open source innovation has a powerful new champion. MiniMax M2 is a fully open source AI model that places world-class intelligence within reach for a wider community of developers, researchers, and organisations. With weights freely available on Hugging Face and flexible licensing under Apache 2.0 and MIT, it allows unrestricted commercial use and customisation. A free trial on the MiniMax platform runs until 7 November 2025, encouraging hands-on adoption and experimentation.
The model is supported by Chinese tech giants Alibaba and Tencent, positioning it as a strong alternative to closed-source heavyweights like GPT-5 and Claude Sonnet 4.5.
MiniMax M2’s sparse Mixture-of-Experts architecture delivers a standout balance of power and efficiency. While the model contains 230 billion parameters, only 10 billion are activated during inference, resulting in lower GPU requirements, reduced latency, and a significantly lower cost of operation.
Independent benchmarks rank MiniMax M2 among the top five AI models globally, delivering competitive performance at a fraction of the cost typically associated with proprietary systems.
The model is engineered for complex task execution, excelling in multi-file code editing, agentic task execution, natural language reasoning, mathematics, science, and technical workflows. This versatility translates to valuable real-world applications, from software development and workflow automation to research support and intelligent agent deployment.
