MiniMax Launches Self-Evolving AI Model M2.7

Chinese AI company MiniMax has unveiled M2.7, a self-evolving AI model built for coding, agentic workflows, and specialized productivity. The model's most distinctive feature is its ability to improve its own development and operational systems, rather than simply assisting users with tasks. With a context window of approximately 200,000 tokens, M2.7 can process lengthy instructions, large codebases, and research notes in a single session. Performance benchmarks are impressive: 56.22% on SWE-Pro, 55.6% on VIBE-Pro, and 57.0% on Terminal Bench 2. The self-improvement process works through the model operating within a research agent harness and iteratively refining that harness itself. M2.7 can autonomously handle approximately 30-50% of this workflow, allowing human researchers to focus on higher-level decision-making. In demonstration experiments, the model executed over 100 rounds of programming scaffold optimization without human intervention, achieving a 30% performance improvement. In machine learning competitions, it achieved an average medal rate of 66.6% during 24-hour autonomous runs. The GDPval AA ELO score stands at 1495, and the model scored 46.3% on Toolathon. M2.7 represents a significant step toward making AI self-evolution a practical reality, potentially transforming how AI models are developed and refined. The model challenges the traditional paradigm where AI improvement relies primarily on human-directed retraining, opening new possibilities for autonomous AI development cycles.