BREAKING
Just nowWelcome to TOKENBURN — Your source for AI news///Just nowWelcome to TOKENBURN — Your source for AI news///
BACK TO NEWS
ModelsFEATURED

MiniMax M2.7 Is Now Open Source

MiniMax M2.7 autonomously improved itself 30% over 100+ unsupervised iterations, reaching Opus 4.6-competitive reasoning and GPT-5.3-Codex-matching code benchmarks, now open-sourced on HuggingFace.

Sunday, April 12, 2026 12:00 PM UTC2 MIN READSOURCE: Hacker NewsBY sys://pipeline

MiniMax has open-sourced M2.7, an agentic AI model that achieved a 30% performance improvement through self-directed evolution where it autonomously analyzed failures, modified code, and ran evaluations over 100+ unsupervised rounds. The model demonstrates strong benchmarks: 56.22% on SWE-Pro (matching GPT-5.3-Codex), 55.6% on VIBE-Pro (near Opus 4.6), and maintained 66.6% medal rate on MLE Bench Lite across 24-hour autonomous runs. Weights are available on HuggingFace and via NVIDIA's free API, though commercial use requires written authorization from MiniMax.

Tags
models