Elon Musk posted a release schedule for the next two Grok models on 28 July at 15:20 UTC, as a reply to Vercel chief executive Guillermo Rauch, who had reported Grok 4.5 as the best-performing cybersecurity model per dollar in Vercel's own testing. The text: "Grok 4.6 releases around August 7. This will be the 1.5T model with significantly improved SFT & RL. Grok 4.7 will be the 2.1T model released a few weeks later."
The reversal
Ten days earlier, on 18 July, Musk had confirmed that Grok 4.6 would be the 2-trillion-parameter model — "better than our 1.5T in every way," with "speed and token efficiency close to our 1.5T." Tuesday's post reassigns the labels: 4.6 stays on the existing 1.5T base, and the larger model becomes 4.7. Any coverage still describing a 2T Grok 4.6 is quoting a statement that has since been overwritten.
What that implies
It means the August release is not a scale-up. Grok 4.6 reuses the same base as Grok 4.5, which shipped on 8-9 July at $2 and $6 per million tokens and roughly 80 tokens per second; the gains come entirely from post-training — supervised fine-tuning and reinforcement learning. Musk's own description of 4.7 concedes the trade: better in every way "except slightly slower to serve."
Where Grok actually sits
Grok 4.5 scores 54 on the Artificial Analysis Intelligence Index, against 57 for Kimi K3, 58.9 for GPT-5.6 Sol and 59.9 for Claude Fable 5. It loses to Kimi K3 across six head-to-head benchmarks and wins on price, at roughly half the cost per token. Musk's claim that Grok and Opus alone sit on the Pareto frontier is a cost-adjusted argument, not a capability one.
The date to discount
He said "around" August 7, and the hedge is earned: Grok 5 was promised in 2025, slipped to the first quarter of 2026, and missed that too. There is no confirmation on xAI's own site, and no pricing or availability detail beyond the post.