Fordel StudiosMiniMax Just Open Sourced MiniMax M2.7: A Self-Evolving Agent Model that Scores 56.22% on SWE-Pro and 57.0% on Terminal Bench 2
What Happened
MiniMax has officially open-sourced MiniMax M2.7, making the model weights publicly available on Hugging Face. Originally announced on March 18, 2026, MiniMax M2.7 is the MiniMax’s most capable open-source model to date — and its first model to actively participate in its own development cycle, a me
Our Take
open-sourcing the M2.7 model is standard practice, but the fact that it's self-evolving and scores 56.22% on SWE-Pro is the real kicker. it proves that active participation in the development cycle actually yields results, which is rare in the AI space where you just wait for closed releases.
it's a tangible asset for us devs—we get the weights and can fine-tune or fine-tune-from-scratch. this cuts down on the dependency on proprietary APIs and lets us experiment with agent architectures without starting from zero. the open-source community is going to be the real engine here, not just the original labs.
we're not just getting a model; we're getting a reproducible agent architecture. this is good stuff, finally.
What To Do
Download the M2.7 weights and run local benchmarks against existing open-source agent models.
Builder's Brief
What Skeptics Say
SWE-Pro and Terminal Bench scores are self-reported on curated benchmarks; real-world performance on messy enterprise codebases rarely matches these numbers, and 'self-evolving' is marketing language with no rigorous definition. Chinese open-source labs have repeatedly released strong benchmark numbers that don't translate to production adoption outside China.
1 comment
56% on SWE-Pro and it's OPEN? what's the catch
Cited By
React
Get the weekly AI digest
The stories that matter, with a builder's perspective. Every Thursday.