RyanLee

Self-hosted M2.7 for code writing is absolutely allowed and free of charge!!!😁 I think this license is not detailed enough, so I will update it.

@OmarBessa I understand your concerns very well. In reality, we have no way of knowing whether it is being used internally within a company unless it is being sold as an external service. So I don’t think this is an issue, as long as it is not offered as a service to the public.

@RyanLeeMiniMax I'm saying this because I have a framework whose output can be interpreted as distillation. However, it is only for consumer customization. Although I'm considering rolling it at enterprise scale. I'm guessing in this case it would not be allowed without a special license.

@OmarBessa Distillation ultimately depends on its intended purpose. If the goal of distillation is large-scale commercial use, like in the case of Kosher, it is definitely need a license. However, if distillation is only for personal coding and personal workflow, we fully support it.

@RyanLeeMiniMax What about distillation? Let's say I use your models to train one of my own.

@Eryx_R3D As long as it is not a for-profit product for external use, it does not count as "commercial".

@RyanLeeMiniMax Thank you for your hard work. 🫠Just to double-check, and I mean no offense, would there be a fee if we use this model as a base for our company's workflow?

@MiniMax_AI x.com/RyanLeeMiniMax…

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：huggingface.co/MiniMaxAI/Mini… Blog: minimax.io/news/minimax-m… MiniMax API: platform.minimax.io

@oneitonitram Good idea!

@RyanLeeMiniMax You could allow all providers offer the model, without your license(agreement) but provide a list of authorised providers, for those you have worked with to provider quality models. win win for everyone, if the fear is quantized, substandard versions

@rodrimora Nope, don't require

@RyanLeeMiniMax But it does require to show "Built with MiniMax-M2.7" permanently, right?

@ronaldmannak True

Good overview. I know it’s controversial but I think it’s reasonable and am glad MiniMax decided to open source the weights. And in case people wonder: I understand the commercial license is currently only available to inference providers, not local

@jeanlouisug Updated

@RyanLeeMiniMax Yo Ryan, Your Hugging Face page says "open source." It's not. It's non-commercial + ask permission. That's proprietary/ Feed your own model your license and ask: "Is this open source?" It'll tell you no. So stop pretending and lying. Remove "open source" from your HF page.

This is my first time judging a hackathon at YC. This robot-themed hackathon is really creative Michael is a 17-year-old high school student, but he won an award in this hackathon. Congratulations!

@darekgusto Thanks Darek, We will keep going

I root for you, I want you to succeed and progress. Your models are mighty awesome, your pricing plans are well-structured and the limits generous. And I do appreciate a personal take a lot. But for many, it will still read like PR one. Mostly because you focus on one argument that's easier for the audience to feel positively about, and that's what PR departments do. Sure, that's a good argument, and important for people to learn about. It's good you explained it, a lot of us don't know such stuff happens. Still, the sustainability argument is also strong and valid. It is important for the AI labs to make monies. And it was a good chance to shine more light on that, but it got one short paragraph, which again, makes it feel like PR. And come on, we all know Kimi K2.5 and Composer 2 situation should have some influence on that. Although I totally understand that you and Minimax as a lab don't necessarily want to discuss other companies' dramas. Keep up good job, M2.7 is a great model!

@The_ERP_Man Self-hosted M2.7 for code writing is absolutely allowed and free of charge!!😁 I think this license is not detailed enough, so I will update it.

@RyanLeeMiniMax Soundd logical. However - If I use MiniMax in my coding assistant to write code, and I sell that resulting code independently, do you consider that commercial use? That is what is still not clear. Thanks.

@YouJiacheng Self-hosted M2.7 for code writing is absolutely allowed and free of charge!!😁 I think this license is not detailed enough, I will update it.

based on the license, you can't even use a self-hosted M2.7 to write code for your commerical apps.

@xeophon Thanks, It's better named "open weight", I will change

@RyanLeeMiniMax 1. Naming the release "open source" isn't true, hopefully you update the official wording if you decide to keep those licenses 2. People are somewhat confused what "derivatives" entail, as its not described. Gemma3 was more explicit what derivatives are x.com/YouJiacheng/st…

@Yuchenj_UW You get the point and get more clear Image. 🤣

MiniMax M2.7 is open-source! The most interesting part of its blog is the “model’s self-evolution.” It’s essentially Karpathy’s Autoresearch. They did two things: 1. They built a research agent to collaborate with their RL team: “A researcher starts by discussing an experimental idea with the agent, which helps with literature review, tracks a pre-set experiment spec, pipelines data and other artifacts, and launches experiments. During the experiments, the agent monitors and profiles progress and automatically triggers log reading, debugging, metric analysis, code fixes, merge requests, and smoke tests, identifying and configuring subtle yet key changes.” “M2.7 is capable of handling 30%–50% of the workflow.” 2. The model rewrites its own harness: MiniMax had M2.7 improve coding performance on an internal scaffold, running a fully automated loop for 100+ rounds. Internal evals improved by 30%. I'm pretty sure every AI lab is doing Autoresearch in some way internally.

@Xianbao_QIAN Thanks for understanding Tiezhen. x.com/RyanLeeMiniMax…

a new paradigm

@9hills x.com/RyanLeeMiniMax… 感谢理解，我们一直在开源和可持续发展上寻找平衡，非盈利目的是使用是非常鼓励和欢迎的。

中国开源旗舰级模型可商用时代结束。 随着火山、阿里、百度等云厂商提供开源模型（GLM、Kimi、MiniMax）coding plan 以来，模型服务商的商业空间被压缩，可以预言后续即使是开源，也不允许商用了。 开源的市场认知建设完毕，付费认知已被广泛接受。

@0xSero Thanks for understanding! x.com/RyanLeeMiniMax…

Defending MiniMax I think it’s fine they set up a license. We kind of deserve it. You work on something with dozens of people burning through investment money. Put it out to the world for free. People say you’re benchmaxxing, call you a communist spy, take your models and sell them. Demand more from you. You deal with this for a year, your peer’s models are being used in the west with little to no credits, same story. They released the research, gave us the weights, released the environments, proved you can max 230B params competitive. Thank you MiniMax. I hope you win more and more.

@xeophon x.com/RyanLeeMiniMax… Hello everyone, hope this is helpful

wow, they did a non-commercial license... M2: Display the name if >30M revenue / 100M users M2.1: Display the name M2.5: Acceptable use policy M2.7: Non-Commercial license

@bnjmn_marie x.com/RyanLeeMiniMax… hello Benjamin, for personal use it's absolute free.

Minimax M2.7: - Marketing: "we open-sourced the model" - Reality: released only the weights and you can't use the model if you make money with it