MiniAX released M2.7, a new open-source model that posts highly competitive scores on public coding benchmarks — strong enough to put it in the conversation with the leading frontier models on common evaluation suites.
The catch is the license. MiniAX M2.7's terms strictly prohibit commercial use. Developers, researchers, and hobbyists can download the weights, fine-tune, evaluate, and build non-commercial projects on top of it, but any production deployment tied to revenue is out of scope under the current license. That makes M2.7 a research release and a competitive signal rather than a true open-source alternative for commercial teams.
A few things worth flagging:
Benchmarks vs. usefulness: the coding scores are strong, but the commercial restriction severely limits where M2.7 can actually show up in real products. Teams evaluating it as a potential replacement for paid APIs should read the license carefully before running pilots.
Signaling effect: the benchmark numbers keep the pressure on Western frontier labs and on other open-weight releases. Even if M2.7 is not deployable in a typical SaaS context, its quality at the open tier raises the floor for what any proprietary coding model needs to deliver.
License watch: prior open-weight models from Chinese labs have loosened their licenses over time as successors shipped. Whether M2.7's terms follow that pattern will shape its ultimate impact.
For now, M2.7 is a technically impressive release with a practical ceiling — interesting to study, cautiously usable for internal tools, but not a commercial-grade option out of the box.