MiniMax M2.7
MiniMax M2.7 is a proprietary, self-evolving large language model released in March 2026, designed for building agent harnesses, updating memory, and iteratively improving workflows. It delivers significant performance gains on SWE-bench-style coding tasks.