MiniMax M2.7 Review: Self-Evolving Agentic LLM — License Controversy, Benchmark Regressions, and the Self-Evolution Story

MiniMax M2.7 (March 2026) introduces 'self-evolution' training — the model managed its own RL pipeline — but ships with a restrictive commercial license, a slower inference speed, and a SWE-Bench Verified score that regressed from M2.5. We review the architecture, training innovation, benchmarks, pricing, and whether the 3.5/5 rating is earned.