SAN FRANCISCO — xAI's next foundation model is nearly ready to ship. On May 25, 2026, Elon Musk confirmed that Grok V9 Medium — a 1.5 trillion-parameter model three times larger than the version currently handling all Grok production traffic — has completed its initial training phase, with a public release expected approximately two to three weeks from that announcement. That window places the expected launch in mid-June 2026.
The disclosure is significant for anyone following the competitive AI landscape. Grok's current production backbone, a 500-billion-parameter model internally called v8-small, handles all live queries across the Grok platform. Musk has publicly described the v8-small as "just 0.5T" and acknowledged it lacks important training data. V9 Medium is the correction — and the training data decisions behind it suggest a deliberate push toward developer use cases.
Cursor Data: A Developer-First Training Strategy
The most consequential element of V9 Medium is not its parameter count but its training corpus. xAI reached an agreement with Anysphere, the developer of Cursor — one of the most widely used AI-augmented code editors in the industry — that involved using Cursor's real-world developer workflow data to supplement V9 Medium's training. Unlike models trained primarily on public GitHub repositories, V9 Medium was exposed to how actual developers debug, refactor, and extend production codebases.
Musk noted in his May 25 announcement that more Cursor data will continue to flow into the model as training progresses. In April 2026, SpaceX announced it had secured an option to acquire Anysphere for $60 billion, or alternatively to pay $10 billion for a deep collaborative arrangement. The V9 Medium training partnership gives that deal a clear technical rationale: xAI is building an AI stack in which Cursor and Grok are integrated from the foundation up.
Benchmarks and the Road to Mid-June
Supervised fine-tuning was underway at the time of Musk's announcement, with reinforcement learning — the final phase before deployment — expected to begin within days. V9 Medium has not yet been released to public evaluators, so no independent benchmark data exists. What Musk has stated directly is that the model delivers major improvements on complex programming tasks compared to the current v8-small.
