MiniMax M2.7 is a mixture-of-experts language model designed for complex agentic workflows and production-grade productivity tasks. With a focused 10 billion activated parameters, it achieves strong real-world capability while maintaining low latency and scalability. The model is purpose-built for agent harnesses, supporting multi-agent collaboration through native Agent Teams with role identity and autonomous decision-making across dynamic environments. Its design emphasizes top-tier engineering performance, professional office delivery, and character-rich interaction, making it well-suited for elaborate tasks that require coordinated, long-horizon execution.

What sets M2.7 apart is its meaningful participation in its own development — an internal version autonomously ran over 100 optimization rounds, analyzing failure trajectories, modifying code, evaluating results, and deciding to keep or revert changes. This self-directed reinforcement learning loop delivered a 30% improvement on internal programming benchmarks. The model demonstrates strong software engineering results on SWE-Pro and multilingual benchmarks, handles end-to-end project delivery across Web, Android, and iOS, and supports workflows like live debugging, root cause analysis, financial modeling, and full document generation. With 97% skill compliance across 40+ complex skills and native multi-agent collaboration, M2.7 is positioned as a next-generation model for autonomous real-world productivity and continuous improvement.