MiniMax M2.7

MiniMax M2.7 is a sparse Mixture-of-Experts (MoE) reasoning language model developed by MiniMax, designed for agentic coding, tool use, and multi-step workflow automation. It is positioned for software engineering, document generation, and productivity tasks, with a training process that incorporated self-improvement techniques where the model helped optimize parts of its own development pipeline. Source: supabase.