Chinese AI model MiniMax M2.7 reportedly helped develop itself

2026-03-21 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Advanced, short

Summary

Chinese AI company MiniMax has released M2.7, a new model that reportedly contributed to its own development through autonomous optimization loops, achieving competitive benchmark results. During its development, M2.7 updated its knowledge stores, built complex agent capabilities, and refined its reward-based training process. MiniMax envisions future AI self-evolution transitioning to full autonomy across data construction, model training, inference architecture, and evaluation. M2.7 demonstrated its self-optimization by autonomously improving a model's coding performance over 100 rounds, leading to a 30 percent boost on internal evaluation sets. It also scored 66.6 percent in OpenAI's MLE Bench Lite competitions, comparable to Gemini 3.1, and achieved an ELO score of 1,495 on the GDPval-AA benchmark for office tasks.

Key takeaway

For AI Scientists developing new models, MiniMax M2.7's self-optimization capabilities suggest a viable path toward reducing human involvement in the development lifecycle. You should investigate integrating similar autonomous agent systems into your workflow to handle tasks like debugging and experiment tracking, potentially accelerating model iteration and performance improvements. Consider M2.7's benchmark parity with leading Western models as a reference for your own self-improving AI initiatives.

Key insights

MiniMax M2.7 demonstrates AI self-improvement by autonomously optimizing its own training and development processes.

Principles

• AI agents can manage literature research and debugging.
• Autonomous optimization can boost model performance.
• Benchmark scores are indicators, not definitive real-world measures.

Method

M2.7 uses an internal research agent system to handle tasks like experiment tracking, code fixes, and metric analysis, with human researchers intervening only for critical decisions, covering 30-50% of the workflow.

In practice

• Use M2.7 for financial analysis and report generation.
• Integrate M2.7 via MiniMax Agent or API platform.
• Explore OpenRoom for graphical AI interaction.

Topics

• Self-Improving AI
• Large Language Models
• AI Benchmarking
• Autonomous Agents
• AI Development

Code references

• MiniMax-AI/OpenRoom

Best for: NLP Engineer, AI Scientist, Research Scientist, AI Engineer, Machine Learning Engineer, AI Researcher

Related on AIssential

• this EX-OPENAI RESEARCHER just released it... — AI agents can autonomously conduct machine learning research, improving model training and performance.
• MiniMax M2.7 — The Loop of Progress — MiniMax M2.7 introduces self-evolution, allowing AI to participate in its own development and accelerate progress.
• TAI #195: GPT-5.4 and the Arrival of AI Self-Improvement? — AI is evolving into a closed-loop system capable of autonomously improving its own development stack and operational efficiency.
• MiniMax's new open M2.5 and M2.5 Lightning near state-of-the-art while costing 1/20th of Claude Opus 4.6 — MiniMax's M2.5 models offer near state-of-the-art AI performance at a fraction of the cost, enabling widespread agentic applications.
• New MiniMax M2.7 proprietary AI model is 'self-evolving' and can perform 30-50% of reinforcement learning research workflow — MiniMax M2.7 introduces self-evolving AI capabilities, autonomously managing significant portions of its own development workflow.
• Intelligence with Everyone: RL @ MiniMax, with Olive Song, from AIE NYC & Inference by Turing Post — MiniMax trains open-weight models using RL, expert feedback, and environment perturbations for robust, agentic performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.