Grok V9-Medium completes 1.5T training, release due in 2-3 weeks

xAI’s next Grok model has moved out of its main training run with a concrete size, timeline, and workload target. In a May 25, 2026 post on X, Elon Musk said Grok V9-Medium has finished training at 1.5T parameters and is now heading through fine-tuning and reinforcement learning before public release.

“Grok foundation model V9-Medium (1.5T) has finished training.”

The important comparison is the production baseline. Musk said V9-Medium should be a major improvement over the 0.5T v8-small model that currently serves all Grok production traffic. That makes the new model three times larger by parameter count, while the post also says a lot of Cursor data was added during supplementary training. For developers, that is the most specific clue: xAI appears to be pushing Grok harder toward difficult coding tasks rather than only general chat or search assistance.

Musk’s X account has often acted as the fastest channel for xAI model status, training updates, infrastructure notes, and release windows. This post fits that pattern because it gives the staged path rather than only a finished product claim: main training complete, fine-tuning underway, reinforcement learning starting in a few days, then public release in 2 to 3 weeks. The linked News247 summary also highlighted the same concrete numbers, framing the move as a jump from a 0.5T production model to a 1.5T foundation model.

The next test is evidence outside the training room. Watch whether V9-Medium shows up in coding benchmarks such as SWE-bench-style repair tasks, terminal agents, and long multi-file edits, and whether xAI can serve a larger model without making Grok slower or more expensive for everyday traffic. If the release window holds, the first public signal should arrive in mid-June 2026. The source post is available on Elon Musk’s X account.