Grok 4.7 Release Date: SpaceX Data, a Slipped Timeline, and What xAI Has Actually Said

Grok 4.7 got a new — and later — release window yesterday, and the reason it slipped is the most unusual training-data story in frontier AI. As of August 13, 2026, the x​AI model list at docs.x.ai still tops out at grok-4.6 — meaning the newest Grok you can actually call is Grok 4.6, the flagship x​AI shipped on August 12. There is no grok-4.7 model ID, no pricing, no context window, and no benchmark card. But on his X account the same day, Elon Musk confirmed the model is real and further along than anything x​AI has said about a model before it ships: initial training is complete, and the company is now feeding a "massive amount" of SpaceX engineering data into a supplemental training run. He put the release at "3 to 4 weeks," which points to early September — and which is already a slip from the window he floated in late July.

The one-line answer, if that's all you came for: Grok 4.7 is not released, and the closest thing to an official date is Musk's August 12 "3 to 4 weeks," which lands it around September 2–9. The window he gave on July 25 pointed to roughly August 22; that date is gone. This article separates what is actually confirmed about Grok 4.7 from what is Musk's claim or a community report, walks through the SpaceX data angle that makes this release different, and ends with how to know first when the model is real.

What is actually known about Grok 4.7 (it's a short section)

The confirmed record on Grok 4.7 fits in three lines, because x​AI's own pages say nothing about it yet. First, the model list at docs.x.ai ends at grok-4.6 — no grok-4.7 entry, no price card, no context window, no release note. That was re-verified on August 13. Second, Musk stated on August 12 that "initial training is complete and now we're adding a massive amount of SpaceX company data in supplemental training." That is the strongest single fact about the model: it exists, and it is in its final training phase rather than the rumor stage. Third, he set expectations for what the corpus is meant to buy — "the SpaceX training corpus is so awesome & unique" — and said he would "be shocked if any model is better at real-world engineering than 4.7."

Everything else with a number attached is a claim with a source attached, and the sourcing matters because the whole point of a release-date piece is not to launder a founder's remark into a spec.

Everything with a number is a claim — labeled as such

Here is what is circulating about Grok 4.7 and the honest label for each item.

• A roughly 2.1-trillion-parameter design — reported across multiple outlets and traced to Musk's own remarks, but never placed on any x​AI specification sheet. Treat it as a claim, not a confirmed spec. For scale, the current flagship Grok 4.6 is the 1.5-trillion-parameter model.

• "Better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency" — Musk's description from late July. It is his positioning, and there is nothing to independently verify it against: the model is not running anywhere yet.

• "Grok 4.7 will exceed all current models" — Musk, August 12. A product claim, not a benchmark. No independent score exists for a model that has not shipped.

• "Real-world engineering" supremacy — Musk's framing, tied entirely to the SpaceX corpus. "Real-world engineering" is not defined in the post, and no rerunnable engineering eval has been published, so the claim cannot currently be tested even in principle.

• What the SpaceX corpus contains — reported in detail by tech outlets: Starlink satellite telemetry, rocket development and test data such as Raptor combustion-chamber pressure logs and Starship re-entry telemetry, plus internal engineering artifacts like Slack threads, Jira tickets, GitHub repositories and ERP records. This is community reporting of what Musk has said, not an x​AI disclosure, and ITAR-restricted hardware specifics are expected to be excluded.

The SpaceX data is the story — and the open question

Whatever the release date, the SpaceX training corpus is the part that changes the conversation. The closest precedent is the previous generation: Grok 4.5 was supplemented with Cursor IDE traces, a code-usage corpus, in a late training run. Grok 4.7 goes further in kind — rocket telemetry, manufacturing records, CAD and failure logs are not code. If the claims hold, x​AI would be giving a frontier model something no rival has: a couple of decades of real engineering operations, including the failures, and an intuition for physical and manufacturing constraints that text-only corpora never contain.

That is the bull case, and it is also the honest boundary of what we know. None of the details have been disclosed — no token counts, no corpus mix, no license or export-control filtering, no engineering evaluation. "So awesome & unique" is a claim, and claims need a lab. The test will come when the model ships and independent evaluators can run it against real engineering tasks; until then the SpaceX story is the most interesting unverifiable thing in frontier AI, and the discipline is to call it exactly that.

The timeline, and the pattern you should notice

Grok 4.7 has had two official-ish windows, and the second is later than the first. On July 25 Musk said the model was about four weeks out, which implied a release around August 22. On August 12 he revised that to "3 to 4 weeks," which implies September 2–9. That is a two-to-three-week slip, and the August 22 window is gone.

The pattern is the part worth internalizing, because x​AI's verbal deadlines have a measured track record. Musk's July framing pointed at Grok 4.6 around August 7; Grok 4.6 actually shipped August 12. A release window from x​AI is a statement of intent, not a commitment — the honest reading of "3 to 4 weeks" is "Grok 4.7 is real and still coming," with a strong prior that the date drifts toward the later end.

Where Grok 4.7 sits relative to Grok 4.6

To reason about what Grok 4.7 must beat, anchor on the model x​AI actually shipped. Grok 4.6 launched August 12: $2 per million input tokens and $6 per million output below 200K input (doubling to $4/$12 above it), a 500,000-token context window, a February 1, 2026 knowledge cutoff, and an Artificial Analysis Intelligence Index of 61. Grok 4.5, released in July, sits below it at 56.

• Parameters — ~2.1T (reported, unconfirmed) for Grok 4.7 versus 1.5T for Grok 4.6.

• Status — in supplemental training versus released August 12, 2026.

• Price per 1M tokens — none published versus $2 in / $6 out.

• Context window — none published versus 500K tokens.

• Serving speed — Musk says slightly slower than Grok 4.6 versus today's ceiling.

• Token efficiency — Musk says better than Grok 4.6 versus the current baseline.

• Independent score — none versus an AA Intelligence Index of 61.

And the field Grok 4.7 is aimed at, on the same scoreboard: Claude Fable 5 at 62, Claude Opus 5 at 63, and Kimi K3 at 60 — with GPT-5.6 Sol tied with Grok 4.6 at 61. Musk did not frame this as beating a single rival; in the same post that claimed Grok 4.7 "will exceed all current models," he acknowledged that "Anthropic is a great company and will probably release improved models soon." The competitive reading is that x​AI sees Claude Fable 5, Claude Opus 5, and Kimi K3 as the models it has to pass.

What this means for teams calling Grok today

For developers, the practical ceiling on the API is Grok 4.6 — the model you call is exactly that, not the one in the teasers. On OrcaRouter's catalog the Grok series currently runs Grok 4.6 and Grok 4.5, billed at the provider's list price with 0% markup, so the $2/$6 you put in your cost model is the $2/$6 on your invoice. When Grok 4.7 ships and a provider onboards it, we'll list it the same way at list price — one key, no new integration, no second contract — but that day is the model list changing, not today.

That same routing layer is the low-risk way to move when Grok 4.7 does land, and the SpaceX story makes the caution more pointed rather than less. A brand-new model's scores are vendor claims until independent labs reproduce them — the first-week numbers are precisely the ones to distrust — and automatic failover lets you point traffic at it without betting a production path on a vendor card. If a regression shows up on real requests, the next request falls back to a provider that still answers. That is the difference between trying an unproven model and betting your pipeline on it.

How to know first (better than the rumor mill)

Everything above goes stale the day x​AI flips a switch, so here is the check that beats every tracker:

• Watch the model list, not the tweets. The moment Grok 4.7 is real, docs.x.ai/docs/models gains a grok-4.7 model ID with pricing and a context window attached. That is how Grok 4.6 appeared — the model list first, the announcement after.

• x​AI's release notes and its X account are the vendor's own channels, but the model list is the artifact, because it is what the API actually serves. A model ID on the docs page is a fact; a mention on X is a statement.

• Watch the OrcaRouter model catalog. A grok-4.7 model page on orcarouter.ai is the "you can call it today" signal — when a provider has onboarded it and we're listing it, it is routable through one API at list price.

• For consumer-side visibility, the Grok app gains the model as an entry point. API availability and app availability can diverge, so which signal you watch depends on whether you're building or just curious.

The state of play

Grok 4.7 is a real model in its final training phase, and it is not released, has no spec sheet, and has a verbal window that has already slipped once. The SpaceX engineering corpus genuinely separates this release from anything a rival has attempted — nobody else is feeding rocket telemetry, manufacturing records and failure logs into a frontier model — but "so awesome & unique" is a claim until independent labs reproduce it. If you came here for a release date, the honest answer is "early September, per Musk," with a strong prior, based on how Grok 4.5 and Grok 4.6 actually shipped, that the verbal timeline slips before it lands. The winning move for anyone building today is not to wait — it's to build on what's live at $2/$6 and switch the moment the model list changes. When Grok 4.7 ships, it should land in your key, not in your calendar.