xAI plans to Open-Source Grok-3

xAI, the AI company founded by Elon Musk, has confirmed plans to open-source its upcoming Grok-3 model. This news came directly from Musk when he replied “Yes” to a question on X asking if xAI would open-source Grok-3. However, further details or a specific date are unconfirmed.

Open-sourcing means sharing the model’s full code, weights, and structure with the public. Anyone can then download, study, modify, or build on it for free. This approach supports transparency and lets developers around the world contribute to AI progress. xAI has followed this path before.

In March 2024, the company released Grok-1, a large 314-billion-parameter model, under the open Apache 2.0 license. Developers could access its weights and architecture on GitHub right away. This move allowed researchers and companies to experiment with the base model and create new tools.

Later versions took a different direction. Grok-1.5 arrived in March 2024 with better reasoning and longer context handling. Grok-1.5 Vision was followed in April as the company’s first multimodal model, able to process images. Then, in August, xAI launched Grok-2 and a smaller Grok-2 mini in beta.

These models brought stronger performance in chat, coding, and reasoning. In August 205, the company open-sourced Grok-2 via Huggingface. The company kept its weights and full details private, making them available only through the X platform and API.

The decision to open-source Grok-3 marks a return to the earlier approach. The model is expected to be xAI’s most advanced model yet, with training planned on large computing resources. By sharing it openly, xAI aims to speed up innovation and let the wider community improve the technology.

This choice stands out in the AI field. Many leading companies keep their latest models closed to protect safety and business interests. xAI’s plan for Grok-3 shows a belief that open access can drive faster, broader progress while still addressing risks.

(source)