Previously, Elon Musk announced that Grok 3 training was complete and the model possessed superior computational capabilities.

xAI revealed that it used 200,000 Nvidia H100 GPUs at a data center in Memphis to train the model. The company also said that along with Google Gemini and ChatGPT 4-o, Grok 3 is considered a competitor to DeepSeek- V3 and Claude 3.5 Sonnet.

Grok 3 includes several sub-models, including Grok 3 Reasoning and Grok 3 mini. Grok 3 mini is designed to be more responsive although it may sacrifice some accuracy.

xAI has also run some tests on Grok and GPT-4o. The results show that Grok 3 outperformed GPT-4o on some PhD-level physics and biology questions. The model has already begun to be deployed and has achieved competitive scores in Chatbot Arena, where users can vote for the model they think is the most accurate.