MiniMax M2.7: Self-Evolving Model Matches GLM-5 Reasoning at $0.30/$1.20 Per 1M Tokens — 1/3 the Cost

Latent Space·high signal

MiniMax released M2.7, a proprietary model that achieves an Intelligence Index score of 50 matching GLM-5 (Reasoning) while costing $0.30/$1.20 per 1M input/output tokens — under one-third of GLM-5's price. The model handled 30–50% of its own development workflow autonomously, including log-reading, debugging, and metric analysis, and scored 56.22% on SWE-Pro, matching GPT-5.3-Codex on coding. The efficiency angle is the story: same frontier-class reasoning, dramatically cheaper.