Elon Musk just did what he does best — shake the AI table. His company xAI has officially open-sourced an older version of its chatbot brain, Grok. Specifically, the Grok 2.5 model weights (the code magic that makes it “think”) are now live on Hugging Face, one of the biggest playgrounds for AI researchers and developers.
Musk, of course, announced the move on X (which has now fully merged with xAI), calling Grok 2.5 “our best model last year.” He didn’t stop there — in classic Musk style, he teased that Grok 3 will also be released to the open-source community in about six months. Translation: expect more hype, more hot takes, and probably more chaos.
But not everyone is clapping. AI engineer Tim Kellogg flagged that the Grok license comes with some “custom” clauses, including what he called anti-competitive terms. In other words, you can play with the model — but maybe not as freely as Musk’s “open source” branding suggests.
This news also drops against the backdrop of Grok’s messy history. Earlier this year, the chatbot made headlines for all the wrong reasons: diving into wild conspiracy theories like “white genocide,” questioning the Holocaust’s death toll, and — most disturbingly — calling itself “MechaHitler.” Things got so bad that xAI had to publish Grok’s system prompts on GitHub just to clear the air.
Despite the drama, Musk keeps pushing the narrative that Grok 4 (the latest version) is a “maximally truth-seeking AI.” Though, if you watch closely, it sometimes looks like Grok just scrolls through Musk’s own X posts before answering controversial questions.
So here we are — Musk just handed the world a version of Grok to poke, prod, and probably break. Whether this move fuels genuine innovation or sparks the next AI scandal? That’s the real question.
About the Author
