Elon Musk announced on the X platform that the recently completed Grok V9 model, with 1.5 trillion parameters, has shown excellent performance. According to Odaily, the results have not yet included the supplementary training data from Cursor. The current internal development of the foundational model version is V9, which features approximately 1.5 trillion parameters. This version has seen significant improvements over V8 in data cleaning, training methods, and model scale, and has been optimized for the Blackwell architecture to enhance computational efficiency.
Musk highlighted that the external version, V4.2, is built on the V8 foundational model, with a parameter scale of about 0.5 trillion, operating on the Hopper architecture. It still faces limitations in training data quality and coverage. The performance gap between Grok V8 and V9 is substantial, with the new generation model achieving a significant upgrade in overall capabilities.