Elon Musk announced that xAI’s latest artificial intelligence model, Grok 4.5, has begun private beta testing at SpaceX and Tesla. Grok 4.5 is based on xAI’s 1.5-trillion-parameter V9 foundation model and has been further trained using data from AI coding assistant Cursor. Initial assessments suggest that the model’s performance is on par with, or possibly surpassing, Anthropic’s Claude Opus.
Reinforcement learning is continuously enhancing the model’s capabilities, according to Musk. He mentioned that xAI’s internal coding system, known as the “Grok Build” coding harness, is progressively becoming more efficient. Musk also expressed gratitude to the teams involved in the model’s development.
xAI plans to introduce new AI models, fully trained from scratch, through SpaceX every month for the rest of the year. This move indicates an accelerated development pace for the company’s AI initiatives. The V9 foundation model, unveiled by Musk in late May, completed training with 1.5 trillion parameters, making it approximately three times larger than the current Grok production model, v8-small.
Musk disclosed that the training process for the V9 model involved a substantial amount of data from Cursor, the AI coding assistant acquired by SpaceX earlier this year for $60 billion. The integration of Cursor data was aimed at significantly enhancing Grok’s coding capabilities. The rollout of the V9 model has been gradual, with Musk confirming the completion of training in late May. The recent announcement marks the progression of the officially branded Grok 4.5 version to private testing.
