MiniMax-M2.7
MiniMax-M2.7 offers the same top-tier intelligence as the standard versionโincluding recursive self-evolution and expert-level office productivityโbut is designed for applications requiring sub-second latency and high-speed token generation. Leveraging an enhanced inference backbone architecture, its output speed is 66% faster than the standard model (reaching 100 tps). It is the preferred choice for interactive programming assistants, real-time agent loop execution, and high-throughput enterprise pipelines with stringent completion time requirements.
MiniMax M2.5
MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon the coding expertise of M2.1 to extend into general office work, reaching fluency in generating and operating Word, Excel, and Powerpoint files, context switching between diverse software environments, and working across different agent and human teams.
MiniMax M2.1
MiniMax M2.1
MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks
minimax-m2
minimax-m2
minimax-m2 is a compact and efficient large language model, optimized for end-to-end programming and agent workflows, with 10 billion active parameters (230 billion total parameters), performing near state-of-the-art in general Inference, Tool use, and multi-step task execution, while maintaining low latency and high deployment efficiency. The model excels in code generation, multi-file editing, compile-run-fix loops, and defect repair in test verification, achieving excellent results in benchmarks such as SWE-Bench Verified, Multi-SWE-Bench, and Terminal-Bench, and demonstrating competitiveness in long-cycle task planning, information retrieval, and execution error recovery in agent evaluations like BrowseComp and GAIA. Rated by Artificial Analysis, MiniMax-M2 ranks in the top tier of open-source models in comprehensive intelligence areas such as mathematics, scientific Inference, and instruction following. Its small active parameter count enables fast Inference, high concurrency, and better unit economics, making it ideal for large-scale agent deployment, developer auxiliary Tools, and Inference-driven applications requiring response speed and cost efficiency.