The M2.7 release is a deliberate provocation aimed at the benchmark leaderboard. Weights available on Hugging Face, integrated into NVIDIA NIM for agentic workflows, and carrying benchmark numbers that hold up across a 45-model field — MiniMax has constructed a release designed to be cited in the same breath as OpenAI's coding agents. The 'self-evolving' framing compounds this: it borrows the language proprietary labs use to justify closed training pipelines and turns it into an open-weight credential. What this framing cannot accommodate is the practitioner who does not care who…