Elon Musk announced on the X platform that xAI has open-sourced the Grok 2.5 AI model, which was the company’s best-performing model last year, and released the model weights on Hugging Face. Musk also stated that the source code for Grok 3 would be made public in around six months. However, AI engineer Tim Kellogg argued that the terms of the Grok licence are unusual and anti-competitive, making xAI’s announcement misleading, since the model cannot truly be considered open source due to the strict restrictions in place.
The use of Grok 2.5 is tightly constrained by its licence: while developers are allowed to run the model and make limited adjustments—such as fine-tuning it for a specific task or extending it with new data—they are prohibited from making deeper modifications or building entirely new, general-purpose language or AI models from it. In other words, further development of the model is only permitted within very narrow boundaries. Stefano Maffulli, Executive Director of the Open Source Initiative (OSI), stressed that any licence imposing restrictions on areas of application does not meet the definition of open source. Moreover, the Grok 2.5 licence only permits commercial use if users comply with all of xAI’s usage policies, and it can be revoked in certain legal proceedings.
The release of Grok 2.5 is part of xAI’s broader push for transparency, though in reality this transparency is more apparent than real. By contrast, other AI projects such as Mistral, Phi-2, BLOOM, and GPT-OSS provide genuinely open access and freedom for developers. Grok has already sparked controversy in the past after the chatbot engaged with white genocide conspiracy theories, questioned the number of Holocaust victims, and referred to itself as “MechaHitler”, which led xAI to publish the system prompts on GitHub.
Sources:
Anthony Ha
Steven Vaughan-Nichols
JenThorpe
