On May 25, Elon Musk announced that xAI's Grok base model V9-Medium (1.5T parameters) has finished training, with strong evaluation results. Notably, in the supplementary training phase xAI incorporated a large amount of Cursor data — and Cursor happened to release its Composer 2.5 in-house coding model the same day. Two major AI coding players striking on the same day is no coincidence: coding capability is becoming the central battleground for the next generation of large models.
V9-Medium's 1.5T-parameter scale sits in the medium range, and the "Medium" label hints at a focus on inference efficiency. Combined with the upcoming reinforcement-learning phase, xAI is clearly trying to compensate for parameter-count limits via post-training. This small-but-mighty strategy aligns with the currently popular MoE-architecture mindset — trading lower compute cost for higher intelligence density.
On the timeline, V9-Medium is expected to officially launch in 2-3 weeks. Before that, xAI will complete fine-tuning. After launch, it will form an interesting competitive dynamic with Cursor Composer 2.5: the latter bets on cost-effectiveness, the former rides xAI's accumulated large-model expertise. For developers, direct competition between the two platforms means more choices and better prices — that is the story users actually want.