SAN FRANCISCO — xAI has crossed a critical milestone in its model development pipeline: Grok V9-Medium has completed training. The 1.5-trillion-parameter model, fine-tuned on real-world developer data from Cursor IDE, is now in the reinforcement learning phase and is expected to reach users in mid-June 2026 — making it one of the most anticipated coding-focused AI releases of the year.
Not Grok 5 — A Different Product Entirely
Coverage of xAI this month has focused overwhelmingly on Grok 5, the flagship 6-trillion-parameter model still in training on Colossus 2. V9-Medium is a separate product built for a different purpose. Where Grok 5 is designed as a frontier reasoning and multimodal powerhouse, V9-Medium targets the practical daily workflows of software developers: editing, refactoring, and understanding code across large, real-world codebases.
The numbers underscore the distinction. V9-Medium's 1.5 trillion parameters place it at roughly three times the scale of xAI's current production model, v8-small, which runs at approximately 500 billion parameters. That's a significant jump — but it's not Grok 5 territory.
The Cursor Advantage
The most distinctive aspect of V9-Medium is its training data. xAI secured a substantial corpus of real usage data from Cursor, the AI-native IDE that counts more than 4 million active developers among its users. Unlike training on static GitHub repositories — which capture finished code — Cursor data includes the full interaction loop: multi-file edits, refactoring cycles, back-and-forth correction sequences where a developer accepts, rejects, or modifies model proposals.
This is the kind of data that captures what developers actually do in an IDE, not just what makes it into version control. If xAI has successfully translated this signal into model behavior, V9-Medium should perform particularly well on the edit-heavy tasks that have historically been underserved by frontier AI models.
Timeline and What "Mid-June" Means
Training completed on May 25, 2026. From there, the path to public release includes post-training alignment, reinforcement learning from human feedback, safety evaluation, infrastructure optimization, and staged rollout. For a model at this scale, xAI has historically completed this pipeline in five to six weeks — placing V9-Medium squarely in the June 15–25 window.
Elon Musk confirmed on X that training had wrapped and that fine-tuning was underway. Evaluation results were described as positive. Whether the mid-June window holds depends largely on the RL alignment phase currently in progress.
