MiniMax Open-Sources M2.7 Self-Evolving Agent Model Scoring 56% on SWE-Pro Benchmark

MiniMax released MiniMax M2.7 as an open-source self-evolving agent model, achieving 56.22% on the SWE-Pro coding benchmark and 57.0% on Terminal Bench 2, competitive scores for an openly available model. The model is notable as the first to actively participate in its own development cycle, representing a paradigm shift in LLM construction where models contribute to their own improvement. Model weights are publicly available on Hugging Face, continuing the platform's role as the primary distribution channel for major open-weight releases.