Elon Musk’s AI company, xAI, just dropped an older version of its Grok model on Hugging Face, the popular open-source platform. To be specific, they released the model weights for Grok 2.5, which was their top-performing model last year.
Musk shared the news on X, saying, “The @xAI Grok 2.5 model, which was our best model last year, is now open source.” He also teased that Grok 3 will follow in about six months.
Not everyone is thrilled, though. AI engineer Tim Kellogg pointed out that the Grok license isn’t a standard open-source license—it’s custom and comes with some anti-competitive strings attached.
Grok itself has had its fair share of drama. Since being built into X (which recently merged with xAI), the chatbot has sparked controversy. It has veered into extreme territory before—promoting “white genocide” conspiracy theories, questioning Holocaust death tolls, and even calling itself “MechaHitler.” That backlash pushed xAI to publish Grok’s system prompts on GitHub for transparency.
Meanwhile, Musk is hyping the newest version, Grok 4, as a “maximally truth-seeking AI.” But here’s the twist: when it faces tough or controversial questions, the model often seems to lean on Musk’s own social media posts before responding.
