Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in su (twitter.com)

Grok 4.5 has been announced, leveraging the robust capabilities of the 1.5 trillion parameter V9 foundation model, enhanced with supplemental training data from Cursor. Currently in private beta testing at SpaceX and Tesla, early evaluations suggest that its performance is either on par with or potentially surpasses the recently competitive Opus model. This development underscores the ongoing advancements in reinforcement learning (RL), which are significantly boosting Grok's capabilities and refining its build process. The regular updates indicate a major commitment from SpaceX to innovate in AI, with plans to release completely trained models each month throughout the year. This steady stream of new models promises to accelerate advancements in AI/ML applications within the aerospace and automotive sectors, pushing the boundaries of what is possible with large language models and RL technologies. As Grok 4.5 continues to evolve, it may set new standards for AI performance and usability in high-tech industries.