PANews June 28 news, Musk posted on X platform stating that the Grok 4.5 version built on the 1.5T V9 f base model has entered private testing phase within SpaceX and Tesla, and Cursor data was incorporated in supplementary training. Early evaluations show the model's performance is close to or may even exceed Opus level. Musk also noted that reinforcement learning (RL) continues to significantly improve model capability, while the Grok Build training framework is undergoing constant iterative optimization. Additionally, Musk revealed that a new model completely trained from scratch by SpaceX will be released continuously at a monthly frequency within this year.