Home / Technology / AI Learns to Improve Itself: MiniMax M2.7 Unveiled
AI Learns to Improve Itself: MiniMax M2.7 Unveiled
19 Mar
Summary
- MiniMax M2.7 autonomously handles 30-50% of its development.
- Model achieves performance parity with Google's Gemini 3.1.
- AI offers high cost efficiency for advanced reasoning tasks.
Chinese AI startup MiniMax has unveiled its latest proprietary LLM, M2.7, a significant advancement in AI development. This model autonomously manages between 30% and 50% of its own development workflow, including data pipelines and training environments. This self-evolutionary capability allows M2.7 to optimize its programming performance through iterative loops, marking a shift towards models actively participating in their own progress.
Performance benchmarks show M2.7 rivaling top-tier models, achieving a 66.6% medal rate in machine learning competitions, matching Google's Gemini 3.1. It demonstrates strong capabilities in software engineering and professional document tasks, with a significantly reduced hallucination rate compared to previous versions. The model also offers substantial cost efficiency, costing less than a third of comparable models for similar intelligence levels.
M2.7 is available via the MiniMax API and platforms, with competitive pricing for input and output tokens. While the model is proprietary, MiniMax continues to support open-source contributions. This development signals a broader trend among Chinese AI startups to pursue proprietary frontier models, mirroring strategies seen from U.S. tech giants like OpenAI and Google.
