WORLD3(@WORLD3_AI):Imagine a model that trains itself. MiniMax M2.7 is now live on RouterLink 🔗 ➡️ Industry's first commercial model to deeply participate in its own training pipeline ➡️100+ autonomous RL iterations → 30% performance gain 🤯 ➡️SWE-bench Pro: 56.22% (matching GPT-5.3-Codex) ➡️MLE-Bench Lite medal rate: 66.6% (on par with Gemini-3.1) The agent economy is entering a new chapter — and we're right at the front of it ⚡ #WORLD3 #WAI #AIagent #RouterLink

2026.04.23 12:30

Imagine a model that trains itself. MiniMax M2.7 is now live on RouterLink 🔗 ➡️ Industry's first commercial model to deeply participate in its own training pipeline ➡️100+ autonomous RL iterations → 30% performance gain 🤯 ➡️SWE-bench Pro: 56.22% (matching GPT-5.3-Codex) ➡️MLE-Bench Lite medal rate: 66.6% (on par with Gemini-3.1) The agent economy is entering a new chapter — and we're right at the front of it ⚡ #WORLD3# #WAI# #AIagent# #RouterLink#

MiniMax (official)@MiniMax_AI

2026.04.12 01:00

We're delighted to announce that MiniMax M2.7 is now officially open source. With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%). You can find it on Hugging Face now. Enjoy!🤗 huggingface： Blog: MiniMax API: