The Grok 4.7 release timeline is a target, not a launch. After SpaceXAI shipped Grok 4.6 on August 12, reporting on Elon Musk’s posts described Grok 4.7 as a larger follow-up, about 2.1 trillion parameters, arriving in three to four weeks. SpaceXAI has not published a Grok 4.7 model card, API identifier, or price.
We already made this mistake in public once. The August 7 window for 4.6 slipped until the company posted the real docs on August 12. The same rule applies here. A founder clock is useful for planning a test harness. It is not a reason to rank a model that does not exist yet.
What is actually on the record
| Claim in circulation | Status on August 13 |
|---|---|
| Grok 4.6 is available | Confirmed by SpaceXAI, identifier grok-4.6 |
| Grok 4.7 follows in 3 to 4 weeks | Founder timing in secondary reports |
| About 2.1 trillion parameters | Reported product claim, not a measured quality score |
| Better than 4.6 except slightly slower | Unverified until a public card exists |
| Supplemental training on SpaceX data | Reported, not independently checked |
SpaceXAI’s current model page recommends Grok 4.6 for coding and does not list 4.7. That is the first-party state. Until it changes, Grok 4.7 stays on the pending benchmark page with no Vibe Coding Index and no rank.
How to prepare without getting burned
- Keep Grok 4.6 on a measured trial. That model is real.
- Write the evaluation set now so you are not inventing tasks on launch morning.
- Do not pre-order capacity or rewrite defaults around 4.7.
- When an identifier appears, record price, context, and effort settings before you trust a chart.
- Wait for the same three independent inputs we used for 4.6 before you talk about rank.
The August 8 piece on how to read a release date is the operating manual. Roadmap, announcement, access, and comparable evaluation are four events. Grok 4.7 is still in the first one.
If three to four weeks holds, the window is early September. If it slips, that is normal. The useful prep is a frozen eval set and a rollback path from Grok 4.6, not a slide that already ranks 4.7 first. Parameter counts do not get a seat on the leaderboard.
Bottom line
Mark late August or early September on a planning calendar if you like. Do not staff a team on Grok 4.7. Staff it on the models that already have identifiers, prices, and scores, and reopen this page when SpaceXAI publishes the real card.