MiniMax Launches Open-Source M2 Model with Superior Speed and Cost Efficiency
China-based artificial intelligence startup MiniMax has officially released MiniMax M2, a new open-source large language model specifically engineered for agent workflows and coding tasks. The company claims that M2 delivers twice the inference speed of Anthropic’s Claude Sonnet 4.5 while costing only 8% of its API price. To encourage adoption, the model is currently available for free use for a limited time. MiniMax reports that M2 demonstrates strong performance in code generation and tool utilization across popular developer platforms such as Claude Code, Cursor, and Cline. In addition to the model release, the company has published the model weights on Hugging Face, providing deployment support for vLLM and SGLang frameworks. Furthermore, MiniMax has introduced its own M2-powered agent platform, which features 'Lightning' and 'Pro' modes designed to handle general and advanced automation tasks respectively. This launch highlights the growing competition in the AI sector, particularly among Chinese startups aiming to challenge established Western models by offering high-performance, cost-effective open-source alternatives for developers and enterprises seeking efficient automation solutions.
Wire timeline
MiniMax Launches Open-Source M2 Model with Superior Speed and Cost Efficiency
China-based artificial intelligence startup MiniMax has officially released MiniMax M2, a new open-source large language model specifically engineered for agent workflows and coding tasks. The company claims that M2 delivers twice the inference speed of Anthropic’s Claude Sonnet 4.5 while costing only 8% of its API price. To encourage adoption, the model is currently available for free use for a limited time. MiniMax reports that M2 demonstrates strong performance in code generation and tool utilization across popular developer platforms such as Claude Code, Cursor, and Cline. In addition to the model release, the company has published the model weights on Hugging Face, providing deployment support for vLLM and SGLang frameworks. Furthermore, MiniMax has introduced its own M2-powered agent platform, which features 'Lightning' and 'Pro' modes designed to handle general and advanced automation tasks respectively. This launch highlights the growing competition in the AI sector, particularly among Chinese startups aiming to challenge established Western models by offering high-performance, cost-effective open-source alternatives for developers and enterprises seeking efficient automation solutions.
TechNode