"Open source" my ass -- they are liars
"Open source" my ass. Since when does "open source" mean you need written permission from some company to use it commercially and slap their logo everywhere? That’s not open source. That’s source available with a leash.
The GNU project and the Free Software Foundation spent decades defining what “free software” actually means — freedom to run, study, modify, and redistribute, including commercially. This MiniMax license violates at least two of those freedoms right out of the gate.
Call it what it is: proprietary code with viewable weights. Stop muddying the term “open source” just because you put a model on HuggingFace. It’s misleading, and honestly, it’s getting tiresome.
If I need a lawyer to ship a product, it’s not open source. Period.
Call it what it is: a multi-million dollar model given to you for free research, and you are complaining about the packaging. 😁
holy shit stop complaining
I had a feeling when they delayed the release of the model on Hugging Face, but don't worry too much. Gemma 3 used a terrible license, so even though it was actually a great model, the attention it got was nowhere near proportional to its performance. That's why Google switched to the Apache license for Gemma 4. So, time will fix this—once they realize that the reputation they don't care about now will cause major trouble for adoption rates in the future.
its different than their past for sure, but why are you complaining about free? this doesn't hurt you in any way. You just complaining to complain
Call it what it is: a multi-million dollar model given to you for free research, and you are complaining about the packaging. 😁
Wrong, it is not given to me. Neither I "have it". That is the point of fake and deceptive statement that they are so called "open source". They are not.
Learn the terminology, and stop falsely representing:
The Open Source Definition – Open Source Initiative
https://opensource.org/osd
its different than their past for sure, but why are you complaining about free? this doesn't hurt you in any way. You just complaining to complain
But you are conflating "free" as in money, or cost, and free as in liberty.
The model is not free software-alike, not open source, the problem is not about them being proprietary. I don't mind, there are many proprietary models.
Problem is in their deceptive advertising it as "open source" while it is not.
If it is multi-million dollar model, then they can also be honest in their advertising, they can pay people to advertise honestly.
Calling it "open source" is an insult to huge free software community.
I think the problem here is the license being "Modified-MIT". I have never seen a closed-source model tagged like that and it's really confusing. They should rename the license to not include "MIT" in its name and make it clear what can and can't be done with this model.
The level of entitlement displayed in this post is staggering.
I think the problem here is the license being "Modified-MIT". I have never seen a closed-source model tagged like that and it's really confusing. They should rename the license to not include "MIT" in its name and make it clear what can and can't be done with this model.
Oh, absolutely. But here's the thing — the real problem isn't just the model author. It's MIT for letting this slide.
'Modified-MIT' is like saying 'slightly pregnant.' You're either MIT — meaning zero fucking restrictions beyond attribution — or you're not. There is no 'modified.' That's just a proprietary license with extra marketing wank.
By allowing people to slap 'MIT' on a license that bans commercial use, adds weird moral clauses, or restricts training — MIT is actively shitting on their own legacy. They built the most permissive license in open source, and now they're watching clowns glue 'except if we don't like you' onto it and still calling it MIT.
It's confusing on purpose. It's brand abuse. And MIT sitting on their hands while this happens just means the name 'MIT license' is becoming meaningless.
So yeah, rename that shit to 'Source-Available With Extra Steps' and stop poisoning the MIT name. Or MIT themselves should step up and say: if you modify it, you don't get to use our name. Period.
MIT is the Massachusetts Institute of Technology, a private research university in Cambridge, Massachusetts. Reputable institution! https://www.mit.edu/
These clowns have zero connection to MIT. No affiliation, no permission, no blessing. They just stole the name "MIT" because it sounds trustworthy and open, then glued their own restrictions onto it.
Wh... What? They didn't... That's not...
Look, the license sucks, but... this isn't an MIT thing. MIT doesn't have control over who uses the license, it's just... who originally put the license out. MIT isn't 'letting this slide' or something. That's not... how this works.
And... Just to be clear, 'If I need a lawyer to ship a product, it’s not open source. Period.' -- If you use GPL in your commercial product you absolutely need a lawyer.
I'm interested in comparing and testing models, not commercial use, so I don't care a huge amount. This model has weird problems anyway, but yeah... the license is crap. But..."stole the name 'MIT'" has some real... misunderstanding too.
The reason we now have free and high-quality Linux for development is precisely because of the GNU and EFF’s almost obsessive pursuit of software freedom back in the day. Admittedly, they faced a lot of criticism, but in the end, the industry made significant progress. Our argument here is similar: some may criticize our “greed” for freedom—go ahead, that’s your freedom of speech, and we have ours. But I hope those who criticize realize the cost of compromising with truly greedy business practices. Don’t tell me you’ve never seen it.
I consider myself quite aware that capital-driven public companies will never turn a blind eye to any possibility that could expand their profits. Therefore, if we don't take a firm stand on the licensing terms of these models, and everyone just thinks, "It works for me, what's the harm?" — coupled with the assumption that DeepSeek won't fire the next shot for us (which is indeed unlikely) — then most of the useful models will adopt similar fake open-source licenses in their next release. Furthermore, by the release after that, these models will become closed-source. What will you say then? "There are no new good open-source models anymore. That's okay — we can just use the official inference services. They're not open-source, but they're very cheap, and I think that's great!" Wake up!
We just need to add same special disclaimer to each repo on GitHub, so if it would be trained on any it - they have to agree with terms of it. Means if it is using in any part GPL constrained code - like Linux, for example - whole model is GPL. Works both ways I guess. Also model must keep track of all licenses they were trained on. Wait a second, but it is already trained on kernel sources, which is GPLv2. Ups
Wh... What? They didn't... That's not...
Look, the license sucks, but... this isn't an MIT thing. MIT doesn't have control over who uses the license, it's just... who originally put the license out. MIT isn't 'letting this slide' or something. That's not... how this works.
The real problem is deceptive labeling, not copyright infringement. MiniMax is calling this "open source" when it transparently is not.
The Open Source Definition (OSI) requires:
- No discrimination against fields of endeavor (commercial use is allowed)
- No restriction on who can use it
- No additional restrictions beyond the standard OSD
This license violates all three. It is a source-available, non-commercial license, not open source.
What should MiniMax have done?
And..Well, gosh, they could have fired up their own incredible,
state-of-the-art, totally-not-just-another-fine-tuned-Llama M2.7 model
— you know, the one they're so proud of that they need to protect it
from those nasty commercial users — and simply asked it to generate a
license.
"Hey M2.7, please write a non-commercial license that doesn't pretend to be MIT."
And you know what? Their super-duper model probably would have spat out something perfectly usable in about 1.2 seconds. Something honest. Something like:
"This is not open source. You can't use this commercially without asking us. Here's a list of things we forbid. Don't call it MIT."
But no. They didn't do that.
Instead, they took the MIT License — one of the most permissive, freedom-respecting, "do-whatever-the-fuck-you-want-as-long-as-you-keep-the-notice" licenses in existence — surgically attached a bunch of "you can't do that, and also ask permission, and also here's a prohibited uses appendix," and then had the audacity to call it "MIT-style terms."
Why?
Because they wanted the credibility of "MIT" without the freedom.
They wanted the smell of open source without the taste. The badge without the burden. The warm fuzzy feeling of saying "we're open!" while keeping both hands firmly around the throat of commercial use.
Using their own model to generate an honest license would have required... honesty. And that's apparently harder than just copy-pasting MIT's homework and scribbling "no commercial lol" in the margins.
So no, they didn't use their super-duper model. They used something far older and far uglier: deception wrapped in a copyright notice.
Bravo. 👏👏👏👏👏
I have reviewed your statement. Putting other issues aside for now, I do think that irregular deployment is a problem, but I don't believe it is serious enough to be resolved through such a self-defeating approach. When I use Openrouter (where I believe most third-party traffic is relayed), newly emerging models often suffer from inconsistent quality. I usually wait for the Exacto variant to incorporate them, or manually filter suitable providers. Generally, providers that support cache hits tend to be more compliant (though there are indeed some that fake cache hits while actually having very low hit rates. Before Openrouter provided analytics, I had to export and analyze usage logs myself, but things seem to have improved now). Therefore, for free usage or similar offers from certain tools, quality degradation should be obvious (but you have directly faced harsher misunderstandings from some people, so your feelings might be different — I don't know).
Apart from that, perhaps you could collaborate with Openrouter to improve these issues. As mentioned above, they are already working in that direction and should have sufficient motivation (though there may be obstacles I'm not aware of — again, I don't know).
In any case, from my personal perspective, for the sake of model quality and reputation, introducing such a licensing policy is too harmful. In the eyes of some people (including myself to some extent), with just the precedent set by this license, Minimax has turned from an open-source champion into a dragon.