Elon Musk announced on May 25 that the Grok foundation model V9-Medium, with 1.5 trillion parameters, has completed training with strong evaluation results. The model incorporated a significant amount of Cursor coding data during supplementary training, with more to follow. Fine-tuning is currently underway and reinforcement learning is set to begin within days, putting the public release timeline at two to three weeks. Musk described it as a major step up from v8-small — the 0.5T model that currently handles all Grok production traffic — and singled out difficult coding tasks as the area of most notable improvement, separately confirming in replies that the new model will be “much better at coding.”
Musk also disclosed that the existing 0.5T v8-small model will be open-sourced before the end of 2026, noting it “should still be quite useful.” The 3x parameter jump from the current production model to V9-Medium, combined with Cursor-sourced coding data, positions xAI’s next release as a direct bid to challenge Anthropic’s Claude and Google’s Gemini in developer-focused AI workflows — a segment that has become one of the most contested battlegrounds among frontier labs.