MiniMax M2.7
FrontierMiniMax•Released on 2026-03-18
MiniMax's self-evolving AI model with breakthrough agent capabilities. Demonstrates 30-50% autonomous RL research workflow. Excels at software engineering (SWE-Pro 56.22%), professional office tasks (GDPval-AA Elo 1495), and complex tool-calling with 97% skill adherence. Features significantly reduced hallucination (34% rate) and 20% fewer tokens than competitors.
Voice of the community
sample 856“M2.7 handled between 30% and 50% of its own development workflow. This is not merely automation of rote tasks; the model optimized its own programming performance.”
“Stop it, I already feel like I'm on cocaine after GPT 5.4, 5.4 mini, Nemotron 4B and Mistral 4 small.”
“MiniMax M2.7 achieves a hallucination rate of 34%, which is lower than the rates of 46% for Claude Sonnet 4.6 and 50% for Gemini 3.1 Pro Preview.”
Core Specs
Scenario Scores
Pros & Cons
Pros
- +Self-evolving RL capabilities (30-50% autonomous workflow)
- +Extremely cheap ($0.30/1M input, $1.20/1M output)
- +Low hallucination rate (34% vs 46% Claude Sonnet 4.6)
- +97% skill adherence on complex tasks
- +20% fewer output tokens than competitors
- +MLE Bench Lite 66.6% medal rate (ties Gemini 3.1)
- +GDPval-AA Elo 1495 (highest among accessible models)
Cons
- −Proprietary model (weights not open source)
- −Less known in Western markets
- −Documentation mainly in Chinese
- −Lower BridgeBench score than M2.5
- −Smaller ecosystem than Claude/GPT