Elon Musk's Grok V9-Medium completes supplementary training, introducing Cursor data to target complex programming scenarios

According to Beating Monitoring, Elon Musk announced that the 1.5 trillion parameter basic model Grok V9-Medium under xAI has completed training and is expected to be released to the public within 2 to 3 weeks. Currently, the model has entered the supervised fine-tuning stage and will begin reinforcement learning training in a few days. The R&D team has incorporated a large amount of Cursor's code data during supplementary training and will continue to feed more samples in the future.

Compared to the v8-small model, which currently supports all production traffic for Grok, the 1.5 trillion parameter Grok V9-Medium will achieve significant upgrades in handling complex programming tasks. Previously, Elon Musk publicly admitted that the v8-small (the external testing version Grok 4.2), with only 0.5 trillion parameters, had serious flaws in data quality, comprehensiveness, and proportion. The trained Grok V9-Medium not only expands its parameter scale to three times that of its predecessor but also has been specifically optimized for Blackwell architecture GPUs.