Elon Musk says xAI has finished training Grok V9-Medium, a 1.5 trillion parameter model, and plans a public release in two to three weeks. He shared the update in a post on X. Fine-tuning is running now, with reinforcement learning due to start within days.
The pitch is coding. xAI fed the model "a lot of Cursor data" during supplementary training, Musk wrote, with more on the way. Cursor is the AI code editor plenty of developers use day to day, so the bet is that learning from real coding sessions sharpens Grok on hard programming tasks.
Musk says "evals look good," though that's his own read and no benchmark numbers came with it. The comparison he's drawing is internal: V9-Medium is roughly three times the size of the 0.5T v8-small that currently runs Grok in production. Bigger usually helps. It doesn't guarantee anything.
V9-Medium is the internal version label, not the name users will see in the app. The Cursor detail has already drawn questions about where the training data came from and whether any private code got swept in. xAI hasn't addressed that. Release window: mid-June, if the fine-tuning and RL stages hold to schedule.
Bottom Line
xAI says Grok V9-Medium runs 1.5 trillion parameters, triple the 0.5T model serving Grok today, with public release targeted in two to three weeks.
Quick Facts
- Model: Grok V9-Medium, 1.5 trillion parameters (company-reported)
- Predecessor: 0.5T v8-small, current production model
- Public release: two to three weeks out, per Musk
- Training status: pretrain done, fine-tuning underway, RL begins in days
- Post date: May 25, 2026 (UTC) on X
