According to Elon Musk on X on May 16, Grok V9 training is complete with 1.5 trillion parameters and demonstrates "very strong" performance. The internal base model version is V9, which features significant improvements over V8 in data quality, training methods, model scale, and is optimized for Blackwell architecture to boost compute efficiency.
By contrast, the current public version v4.2 is built on the V8 base model with approximately 0.5 trillion parameters running on Hopper architecture, with certain limitations in training data quality and coverage. Musk emphasized the performance gap between V8 and V9 represents a "massive" upgrade across overall capabilities.