MiniMax M2.7 Exhibits Early Self-Evolution Capabilities for Agentic Tasks

MiniMax announced M2.7, a new model participating deeply in its own evolution, demonstrating early capabilities for self-improvement. M2.7 can build complex agent harnesses and complete elaborate productivity tasks by leveraging Agent Teams, Skills, and dynamic tool search. For instance, the model updated its own memory and built dozens of complex skills to aid reinforcement learning experiments and improved its learning process based on experiment results. M2.7 achieved impressive performance on benchmarks like SWE-Pro (56.22%) and Terminal Bench 2 (57.0%) for real-world software engineering.