Thank you team for the improved license
Thanks to the team for the improved license. It really seems like you took feedback from the 2.7 release and improved the license for M3. Well done!
Enjoy
coming here for same purpose, thanks for sharing nice model, and keep it up!
Enjoy
Thanks @ryanlee-dev , I know you got a lot of grief in the previous release threads, I think the community really appreciates you being able to work through it with your legal team to land on a vastly improved, community-friendly license.
great Job!
very nice, thank you !
A big thank you. Thank you for being open to the model!
The definition of commercial use is still ambiguous, the examples say one thing and definition another. With a strict interpretation, almost everything needs a permission and displaying of the "Built with MiniMax M3" message. See https://gnu.support/software-freedom-fakers/MiniMax-M2-7-s-MIT-Style-License-Is-a-Misleading-Restriction-That-Bans-Commercial-Use-and-Fails-Free-Software-Standards-124111.html, for more details and a lot of examples. But I have also heard many people claim that it only applies to service providers and using the software as a service, and that using the model to make software, even commercial ones is allowed.
So, what's the correct interpretation? It would also be nice if that ambiguity was removed from the license text itself.
Very nice!
Great Job! FP8 is fine on 8 B200, now just waiting for NVFP4 weights 👀
I try to put in on MLX.
Hope someone can help to upload NVFP4 version
Usually Nvidia or RedHatAI, will publish the NVFP4 2-3 weeks after official release.
Someone already made the request on their repo: https://github.com/NVIDIA/Model-Optimizer/issues/1708
This isn't improved at all, Kimi allows all users to use the product. It just requires to display "Made with Kimi" for commercial (>$20M) uses. Here, any provider will need explicit permission to serve it to users. This is effectively going to restrict the number of providers. Case and point, Kimi K2.6 has over 2 dozen providers on openrouter, Minimax 2.7 less than a dozen.
The only (and non-negligible) improvement is for small and medium entreprises that want to deploy it for their employees. But with 400B parameters, they most likely would need to rely on a cloud provider, which might be prohibited from serving the model.
Just to be clear, I am still very grateful to have one more semi-closed model, it's better than completely closed