Particle.news

SpaceXAI’s Grok V9‑Medium Completes Core Training

The company says the 1.5 trillion‑parameter base model will enter reinforcement‑learning soon with public release expected in early to mid June.

Overview

  • Elon Musk announced that Grok V9‑Medium, a 1.5 trillion‑parameter base model, has finished core training and returned promising evaluation results.
  • The team has added large amounts of Cursor data during supplementary training and plans to continue expanding that dataset before release.
  • Fine‑tuning is already under way and Musk said reinforcement‑learning fine‑tuning will begin within days as the final step before public availability.
  • Musk framed V9‑Medium as a major capability upgrade over the current 0.5T V8‑Small production model, saying it should notably improve performance on hard programming tasks.
  • The announcement follows SpaceX’s acquisition and reorganization of xAI into SpaceXAI and could boost developer tools and coding agents by using Cursor’s real-world coding workflows.