Elon Musk’s xAI has opened up the Grok 2.5 model to developers, researchers and curious tinkerers worldwide. He announced that the source code and model weights for Grok 2.5, the company’s 2024 AI, are now available on Hugging Face. And even more interesting, Grok 3, which was released earlier this year, will follow the same open source path in about 6 months.
Grok 2.5 is a large language model that can answer questions, generate text and do complex reasoning. Unlike Grok 1 which xAI released as a raw base model in March 2024, Grok 2.5 comes with a richer feature set, including improved reasoning and real-time data integration from X. The open source release includes the model’s weights (essentially the numerical parameters that define how it processes and generates responses) and the architecture. Developers can now download these from Hugging Face, a platform that hosts machine learning models, and start playing.
- The Matchbox SpaceX Dragon Spacecraft celebrates this cutting-edge SpaceX vehicle that is helping redefine space travel.
- This premium die-cast spacecraft features realistic details and authentic designs that are true to the Matchbox tradition.
- It comes in special, display-ready packaging for the ultimate in collectability.
xAI’s Community License allows free use for research and non-commercial projects but commercial use is restricted. Companies making over $1M in annual revenue need to negotiate separate terms with xAI and no one can use Grok 2.5’s outputs to train or improve other large scale AI models. This ensures xAI retains control over how its tech impacts the broader AI landscape.
And now we have to look forward to Grok 3 being open sourced in February 2026. Grok 3 was released in February 2025 and has about 10x the compute of Grok 2.5 with features like “Big Brain” for reasoning and “DeepSearch” for research. If xAI follows through, the release of Grok 3’s code could change what’s possible with open source AI especially since Musk is talking about Grok 5 by the end of 2025. This would be a yearly cycle: build a model, refine it, then share it with the world and then get feedback and iterate.
