SAN FRANCISCO — Elon Musk posted a brief but significant update on June 5, confirming a "core model improvement" is underway at xAI — a signal that Grok V9-Medium, the company's most powerful production model yet, is moving from the final stages of supervised fine-tuning toward public deployment. The release is expected in mid-June 2026.
What V9-Medium Is
Grok V9-Medium is the successor to the current v8-small model powering most Grok interactions today. Where v8-small operates at approximately 500 billion parameters, V9-Medium scales to 1.5 trillion parameters — three times larger — a shift that xAI engineers expect to translate into substantially sharper reasoning, more reliable code generation, and stronger performance on multi-step agentic tasks.
Training on the Colossus supercluster in Memphis was confirmed complete by late May 2026. Since then, the model has been undergoing supervised fine-tuning and reinforcement learning from human feedback, the standard post-training process that aligns a raw model with user preferences, safety requirements, and instruction-following behavior. Musk's June 5 post describing a "core model improvement" is consistent with this phase completing and the final pre-deployment adjustments being made.
Worktrees Support Added to Grok Build
The same June 5 update confirmed that Grok now supports Git worktrees — a capability that allows developers to maintain multiple working directories from a single repository simultaneously. The addition arrives via Grok Build 0.1, the dedicated agentic coding model that launched to public beta on May 29, 2026.
Worktrees are a meaningful workflow upgrade for complex software projects. Developers using Grok Build for long-running agentic coding tasks — where the model writes, tests, and iterates on code over many steps — can now maintain separate feature branches without switching directories, reducing merge conflicts and enabling more parallel development within a single Grok session.
