MiniMax M2.1 (minimax.io/news/minimax-m21) is an open-source, large language model developed by MiniMax optimized for software engineering, agentic workflows, and complex instruction following. Built on a ~230B parameter Mixture-of-Experts (MoE) architecture with only ~10B activated parameters per token, M2.1 delivers frontier-grade performance across 10+ programming languages (including Rust, Kotlin, Go, and Objective-C), native iOS/Android and Web development aesthetics, interleaved thinking, and universal agent scaffolding (supporting Claude Code, Cline, and Droid) at low token latency.

Official WebsiteAPICoding Plan