Chinese AI startup MiniMax released M2.7, a proprietary large language model capable of self-evolution. The model automated 30% to 50% of its own development workflow. M2.7 managed its own reinforcement learning processes, including building, monitoring, and optimization. This release marks a milestone in recursive self-improvement for artificial intelligence.
The launch signals a strategic shift for Chinese AI firms from open-source to proprietary models. MiniMax now competes directly with US leaders including Google, OpenAI, and Anthropic. Benchmarks show M2.7 matches the software engineering performance of Gemini 3.1 and Claude Opus 4.6. The company offers the model at a significantly lower price point than its US rivals.