✳ the wire · intel
Musk puts Grok 4.6 at around August 7 and 4.7 a few weeks later — 1.5T then 2.1T
This one is not a leak. Musk posted the roadmap on X on July 28: Grok 4.6 lands around August 7 as the 1.5T-parameter model with improved supervised fine-tuning and RL, and Grok 4.7 follows a few weeks after at 2.1T parameters.
His characterization of 4.7: better than 4.6 in every way, except slightly slower to serve, with better token efficiency despite the larger model. A 2.1T model that is more token-efficient than its 1.5T predecessor is a claim about training, not scale, and it is the specific claim worth testing on day one.
Every wire item is source-verified before it posts and graded: confirmed · strong · reported · rumor. Subscribers see the primary source on each one, the reasoning behind its grade, and the full analysis — plus the launch alerts themselves, which fire the minute a model answers rather than whenever the feed catches up.
