Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Grok 2.5 is now open source. Grok 3 will be open source in about 6 months (twitter.com/elonmusk)
42 points by tosh 43 days ago | hide | past | favorite | 24 comments


On one hand it’s an old, obsolete model.

On the other none of the big-name US labs released their old models, even when obsolete, like gpt 3.5 or claude pre-3.5. All we got are some unusable safetyslop like openai OSS that was worse than existing comparable chinese models on the day they released. Google’s gemma was slightly better, but also hardly exciting.

So… go Elon? It’s not much but it’s something, and who knows, maybe it’ll somehow percolate into other companies to release something that is actually good.

The license is pretty hostile, though.


> none of the big-name US labs released their old models

Hasn't Meta been releasing their models?


I assume this is open weights rather than open source?

Non Twitter link:

https://xcancel.com/elonmusk/status/1959379349322313920


Yeah, open weights but with tons of limitations in the license. https://huggingface.co/xai-org/grok-2/blob/main/LICENSE


Maybe we should have a parallel designation for this like "weights available", because this license is quite restrictive and really not comparable to models with weights released under, e.g., Apache or MIT licenses.


What of those are unreasonable terms? Quickly looking into it, it seems mostly that it is restricted to legal use and the provider has to take some responsibility.


Those limitations mean:

- Not open "source" because source isn't available

- Not "open" weights because there are restrictions on use

This model is weights-available. There's nothing open about it.


Doesn't this apply to most of the "open-weight" models like Llama and Gemma? I remember Llama also having some specific terms targeted at big boi players as well.


Correct. There are just about zero open models out there.


Not open source as it has a non-commercial restrictions, among it's other terms I didn't even bother to read.


Open weights only, open source usually entails that the training material is provided and you could replicate the weights yourself. ITT, we'll never get true e2e libre models without serious legislative overhaul and governance.


Yes, it's just open weights however they do allow commercial use but with stipulations to do no harm, follow the law and not circumvent the guardrails in place. I would worry more about the revokable nature of it more than anything for commercial use.


I’m not surprised, lots of people have stopped buying the model 3. In 6 months they might as well give it away.


Smoke and mirrors...hostile license, old model and it's open weight and not open source. Probably the only reason to do this to Elon to be able to say to the other frontier labs: see, we are open!


Open source those models would enable major lawsuits for copyright violation. And I doubt that they own the rights to the training sets either way.

Open weight models are still useful to extract knowledge from. With much lower cost compared to hitting the API



No, the "Grok Community License Agreement" no way shape or form open source. It's not even close. Just more open-source washing marketing BS.


The main use of this I could see is transparency. I imagine you could test prompts and increase your confidence the released system prompts are actually the ones in use.

Also, assuming they continue to release their old models, I suppose this could be used to get more insight on what caused the MechaHitler incident.

Grok is bad for the world. xAI’s data center is producing a huge amount of pollution in Memphis [0]. Additionally, Elon Musk’s attempts to create an AI that is not “woke” have led to it spewing far-right conspiracy theories [1], although generally for short periods of time before fixes are released.

[0]: https://www.selc.org/news/resistance-against-elon-musks-xai-...

[1]: https://www.theguardian.com/technology/2025/may/14/elon-musk...


It might not be woke but it's just as censored as OpenAI.

OpenAI is "Don't say boobs, some conservative investor might take offence!" Grok is: "Don't say gay, but Hitler is okay"

Both are pretty crap in daily use. Sexuality is a part of life, if you use an AI personally you can't really do without. For work it's ok and that's probably why GPT-5 is so corporate. Useless for personal use.

Grok is useless for me as I'm very pro lgbt and anti nazi.

So yeah what do I use now? Llama3.1 abliterated.

Unfortunately a lot of newer models like phi are trained on synthetic data which is much harder to uncensor because they've never seen any data their makers consider questionable. And those things are very polarised as is American society.

What we need here in Europe is a different mix. Sexual topics (18+ obviously) yes, discrimination no, lgbt yes, fascism no. Maybe mistral can deliver that.


Mistral models are largely along the likes of what you were asking for. However, Grok (any version) absolutely is not a “don’t say gay” model, it talks about sexuality of all forms quite openly and fairly and is happy to produce creative content of any level of explicitness about these topics. It’s the least censored unmodified model I’ve encountered on any topic. People dismiss Grok as a Nazi model based on Musk’s politics without using it themselves.


> based on the headlines they skimmed about Musk’s politics


No need to skim headlines, just skim his Twitter history. Perfect headline material from source.


That was my point. Knowledge based on a headline based on a tweet.


ok, sure




Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: