MiniMax Open-Sources M2 Model for Cost-Effective Agentic AI

MiniMax has officially open-sourced and launched its M2 model, specifically designed for agents and code. The model weights are available on Hugging Face, with support from SGLang and vLLM for deployment. MiniMax M2 is priced at $0.30 per million input tokens, which is 8% of Claude 4.5 Sonnet's price, while offering nearly double the inference speed. The company aims to make agent-related capabilities more accessible with its M2-powered Agent product.