Elon Musk: The gap between Grok V9 and V8 is huge; the V9 training version already shows better performance.
AIMPACT News, May 15 (UTC+8), Elon Musk posted on X platform stating that the latest completed Grok V9 (1.5T parameters) training run "performed very well," and this result has not yet been incorporated into the Cursor data supplementary training. The current internally developed base model version is V9, with approximately 1.5 trillion parameters, significantly improved over V8 in data cleaning, training methods, and model scale, and optimized for the Blackwell architecture to enhance computational efficiency. Musk emphasized that, in comparison, the current external version v4.2 is built on the V8 base model, with about 0.5T parameters, running on the Hopper architecture, and still has certain limitations in training data quality and coverage. The difference between Grok V8 and V9 is