17
votes
xAI has open sourced Grok 2.5
Link information
This data is scraped automatically and may be incorrect.
- Title
- Elon Musk on X (formerly Twitter): "The @xAI Grok 2.5 model, which was our best model last year, is now open source. Grok 3 will be made open source in about 6 months. https://t.co/TXM0wyJKOh / X"
- Authors
- Elon Musk
- Word count
- 37 words
What kinds of specs are required to run a model like this?
So $$$$ even after people work on quantizing it.
Edit: oops, forgot to cite my source. Quote yanked from the hugging face release:
https://huggingface.co/xai-org/grok-2
Yea, "open-source" really doesn't mean much to me when the hardware resources are prohibitively expensive. I think it's cool that I can run gemma3 on my laptop but I'm happy to keep my LLM use localised to my own PC.
To me, talking about open-source (rather than, say, open-weights for example) also feels disingenuous because while it’s there now, it’s not exactly reproducible.
But since sooo much more goes into that, it’d be much harder for anyone to publish. Probably impossible if they want to avoid getting sued into oblivion for publishing exactly which data went into the model’s training, too.
This doesn’t mean I think open-source means “free” or reproducible either, just saying… it’d be nice to have a not-big-player entity be able to come up with a good model on their own, and not just rely on competitors leaking/publishing training strategies sometimes here and there.
Agreed. This feels like a SaaS provider releasing their server as an executable and calling it open source. An executable would arguably be much more useful for reverse engineering, even.
Not to mention that the license seems pretty restrictive as well, including stipulations like: