# Grok 4.7: Release Date, What Musk Has Promised, and What xAI Has Shipped

> Grok 4.7 is unreleased as of Sept 8, 2026. Musk's latest target is about Sept 11. Every dated promise, the 2.1T claim, the SpaceX data, and how his dates have held.

- Author: Nitish Garg, Founder & CEO, CellCog
- Published: 2026-09-08
- Canonical (HTML): https://cellcog.ai/blog/grok-4-7-release-date/
- Section: Guides / Choosing a platform
- Publisher: CellCog (https://cellcog.ai), the AI employee platform. Blog index for agents: https://cellcog.ai/blog/llms.txt

## Key points

- Grok 4.7 is not released as of September 8, 2026. xAI's news page, its model directory, its API release notes and its llms.txt contain no 'Grok 4.7'. The current flagship is Grok 4.6, released August 12.
- Every date in circulation comes from Elon Musk on X, not from xAI. He has given four: 'in 4 weeks' (July 24), 'a few weeks' after 4.6 (July 28), 'ready in 3 to 4 weeks' (August 12) and 'comes out in 10 days' (September 1). The first has passed; the newest points at about September 11.
- The specs people repeat are also Musk's: a 2.1 trillion parameter model, slower to serve than 4.6 but more token-efficient, with SpaceX engineering data added in supplemental training. None of it is in an xAI document, and the 4 to 6 trillion figure some sites quote is a blogger's guess.
- His dates have slipped before. Grok 4.5 was 'hopefully out by late May' and shipped publicly July 16. Grok 4.6 was 'around August 7' and shipped August 12. The July target for 4.7 passed on August 21 without a release.
- There is no credible benchmark leak and no confirmed anonymous checkpoint. The one stealth model people guessed was Grok 4.7, Ox Alpha on OpenRouter, was revealed by Z.ai on August 26 as GLM-5.3-Flash.
- Grok 4.7 is described as a new base model, not a Grok Bot feature. xAI's docs name Grok 4.6 as the default model in Grok Build; the September 3 'Designing Grok Bot' post names no model at all.
- This is a living tracker. When xAI publishes a model page, a price or a launch post, the confirmed facts land here the same day next to a scorecard of every claim above.

## At a glance

- **Is Grok 4.7 released?** No. As of September 8, 2026 there is no xAI announcement, model page, API identifier, price or release-notes entry for Grok 4.7. xAI's newest model is Grok 4.6, released August 12, 2026.
- **When is Grok 4.7 coming out?** xAI has not said. Elon Musk posted on September 1 that it 'comes out in 10 days', which points at about September 11. His earlier targets for the same model were late August and the first week of September.
- **What has Musk said Grok 4.7 is?** A 2.1 trillion parameter model that is 'better than 4.6 in every way, except slightly slower to serve', with 'even better token efficiency', trained further on 'a massive amount of SpaceX company data'. Those are his words on X between July 28 and August 12; xAI has published no spec.
- **What powers Grok Bot and Grok Build today?** xAI's documentation names Grok 4.6 as the default model in Grok Build. The Grok Bot documentation and xAI's September 3 design post do not name a model. Nothing official ties Grok 4.7 to either product yet.

Grok 4.7 has a release date in exactly the sense that Elon Musk has posted one. Four times, in fact, and none of them on x.ai. On July 24, 2026 it was "in 4 weeks." On July 28, "a few weeks" after Grok 4.6. On August 12, the day 4.6 launched, "3 to 4 weeks." On September 1, "Grok 4.7 comes out in 10 days," which points at about September 11. xAI, the company, has published nothing: no launch post, no model page, no price, no benchmark.

As of September 8, 2026, we pulled xAI's [news page](https://x.ai/news), its [model directory](https://docs.x.ai/developers/models), its [API release notes](https://docs.x.ai/developers/release-notes) and its [llms.txt](https://docs.x.ai/llms.txt) directly. None of them contains the string "Grok 4.7." The top of the lineup is [Grok 4.6](https://x.ai/news/grok-4-6), released August 12, priced at $2 per million input tokens and $6 per million output tokens, and named in xAI's docs as the default model of Grok Build.

So this page does what our [Gemini 4](https://cellcog.ai/blog/gemini-4-release-date/), [Astra](https://cellcog.ai/blog/openai-astra-release-date/) and [Qwen 4](https://cellcog.ai/blog/qwen-4-release-date/) trackers do: every claim dated, the founder's posts separated from the company's documents and both separated from rumor, updated the same day the story moves. Grok has a particular wrinkle the others do not. Almost all of the "information" is one person's timeline, so the useful question is not only what he said but how his previous dates held up. We answer that below with the shipping record.

## What we actually know

*Table: Grok 4.7 claims vs verifiable status, September 8, 2026*

| Claim | Status |
|---|---|
| Grok 4.7 exists and is in training | Musk on X, repeatedly since July 24; no xAI document |
| "Initial training is complete" | Musk on X, August 12 |
| 2.1 trillion parameters | Musk on X, July 28; unverified, no xAI spec |
| 4 to 6 trillion parameters | NextBigFuture speculation, July 24; contradicts Musk's own figure |
| SpaceX engineering data in supplemental training | Musk on X, July 21 and August 12; amount and effect unknown |
| Slower to serve than 4.6, better token efficiency | Musk on X, July 28 |
| "Will exceed all current models" | Musk's prediction, August 12 and 15; no results |
| Release date | Musk: "in 10 days" (September 1), so about September 11; xAI: nothing |
| Earlier dates | "4 weeks" from July 24 passed August 21 with no release; "3 to 4 weeks" from August 12 runs out September 9 |
| Benchmark leak | None with files or a named source; circulating "leaks" are unattributed |
| Anonymous checkpoint | None attributed with evidence; Ox Alpha (guessed as 4.7) was GLM-5.3-Flash |
| Pricing, context window, API id, model card | Nothing exists |
| xAI's current top model | Grok 4.6, released August 12, 2026, per x.ai and the docs |

## What Musk has said, word for word

The record is his X account, and it is consistent about the shape of the model and elastic about the date.

**July 21:** "SpaceX's massive corpus of world-class engineering data (excluding material blocked by ITAR) will be added during supplemental training of the 2T run. This will dramatically improve Grok's engineering capabilities." ([source](https://x.com/elonmusk/status/2079446276299465185))

**July 24:** "Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks" ([source](https://x.com/elonmusk/status/2080724087593226311))

**July 28:** "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. This will be better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency." ([source](https://x.com/elonmusk/status/2082123925283041545))

**August 12**, the day Grok 4.6 shipped: "Grok 4.7 is significantly better than 4.6 and should be ready in 3 to 4 weeks. Initial training is complete and now we're adding a massive amount of SpaceX company data in supplemental training. This will be something special." ([source](https://x.com/elonmusk/status/2087604711767896527)) Six minutes later: "Grok 4.7 will exceed all current models. That said, Anthropic is a great company and will probably release improved models soon. However, the SpaceX training corpus is so awesome & unique that I would be shocked if any model is better at real-world engineering than 4.7." ([source](https://x.com/elonmusk/status/2087606260539777263))

**August 15:** "Worth trying Grok 4.6. Fable 5 is smarter overall, but Grok is much faster and lower cost. Grok 4.7 has a good chance of exceeding all current models in intelligence." ([source](https://x.com/elonmusk/status/2088735708693602427))

**August 18:** "Try Grok 4.6! Grok 4.7, which is a major upgrade, is coming soon." ([source](https://x.com/elonmusk/status/2089754307453141424))

**September 1:** "Grok 4.7 comes out in 10 days" ([source](https://x.com/elonmusk/status/2094983639780204846))

Two things stand out. First, the August 15 post is the most candid frame he has offered: 4.6 is the fast, cheap option and Fable 5 is "smarter overall," and 4.7 is the model meant to close that gap. Second, the specific claims are all about size and training data, never about a benchmark, a context window or a price. That is what a model looks like from the inside before the evaluation numbers exist.

## The rumor chain, dated

*Table: The Grok 4.7 chain, dated*

| Date | Event | Source type |
|---|---|---|
| Apr 18 | Musk: Grok 4.4 "early May", Grok 4.5 "hopefully out by late May" | Musk on X |
| Jun 28 | Musk: Grok 4.5 in private beta at SpaceX and Tesla | Musk on X |
| Jul 16 | Grok 4.5 released | xAI, official |
| Jul 21 | Musk: SpaceX data goes into "supplemental training of the 2T run" | Musk on X |
| Jul 24 | Musk: "Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks" | Musk on X |
| Jul 24 | "4 to 6 trillion parameters" floated for 4.7 | NextBigFuture, speculation |
| Jul 28 | Musk: 4.6 "around August 7", 4.7 "the 2.1T model" a few weeks later | Musk on X |
| Aug 12 | Grok 4.6 released; Musk: 4.7 "ready in 3 to 4 weeks", initial training complete | xAI official + Musk |
| Aug 15 | Musk: 4.7 "has a good chance of exceeding all current models" | Musk on X |
| Aug 18 | Musk: 4.7 "coming soon" | Musk on X |
| Aug 20 | Ox Alpha appears on OpenRouter; guessed as Grok 4.7 within a day | Community sighting |
| Aug 21 | The July "4 weeks" target passes with no release | Our read |
| Aug 26 | Ox Alpha revealed as GLM-5.3-Flash | Z.ai, official |
| Sep 1 | Musk: "Grok 4.7 comes out in 10 days" | Musk on X |
| Sep 3 | xAI publishes "Designing Grok Bot"; no model named | xAI, official |
| Sep 8 | x.ai news, model docs, release notes, llms.txt: no "Grok 4.7" | Our own read |

**The parameter count is a founder statement, not a spec.** The 2.1T figure is repeated everywhere because Musk wrote it on July 28, alongside 1.5T for Grok 4.6. It is plausible and it is unverified; xAI has never published a parameter count for any Grok 4 model, and counts for mixture-of-experts models are ambiguous even when a lab does publish them. The 4 to 6 trillion range that some pages carry as fact comes from a [NextBigFuture post on July 24](https://www.nextbigfuture.com/2026/07/spacexai-will-release-grok-4.6-in-2-weeks-and-grok-4.7-in-4-weeks.html) that introduces it with the word "speculation." Musk's own later figure contradicts it.

**The SpaceX data is the interesting claim.** Two posts describe it: the engineering corpus, minus material blocked by ITAR export rules, added in supplemental training, what other labs call mid-training. If true, it is a genuinely unusual dataset, and it explains why Musk keeps saying "real-world engineering" rather than naming a benchmark. It also tells you what to watch on launch day: whether xAI's table leads with coding and agentic rows, as the 4.6 table did, or with something new.

**The benchmark leaks are empty.** We looked for a sheet with files behind it and found none. Posts from late July titled "Grok 4.7 leaks are starting to surface" carry no evaluation files and no source. The Grok numbers that do circulate as leaks are the July 2025 Grok 4 scores. Musk's "will exceed all current models" is a prediction from the person selling the model, and we file it as such.

## How Musk's dates have held

The only calibration data for a Musk release date is his previous release dates. Here are the last two Grok models, from his first stated target to xAI's public release.

*Table: Musk's targets vs xAI's ship dates*

| Model | Musk's first target (date posted) | xAI public release | Gap |
|---|---|---|---|
| Grok 4.5 | "Hopefully out by late May" (April 18) | July 16, 2026 | About 46 days after May 31 |
| Grok 4.6 | "In 2 weeks" (July 24), "around August 7" (July 28) | August 12, 2026 | 5 days |
| Grok 4.7 | "In 4 weeks" (July 24), so about August 21 | Not released as of September 8 | 18 days and counting; latest target about September 11 |

Read fairly, this is a pattern of optimism, not fiction. Grok 4.5 was real and shipped; it was seven weeks later than the first date. Grok 4.6 was real and shipped five days after "around August 7." The honest expectation for 4.7 is the model xAI describes on launch day, on a date somewhat after the one Musk posts. "In 10 days" from September 1 is the tightest window he has given, and it is also the third window for the same model.

## Ox Alpha was not Grok 4.7

The one moment this summer when a Grok 4.7 sighting seemed possible was Ox Alpha. On August 20 an anonymous stealth model appeared on [OpenRouter](https://openrouter.ai/stealth/ox-alpha) with a one million token context and a free window, and by August 21 people on X were guessing it was "Grok-4.7 or Gemini 3.8 Flash." We tracked it as it happened in [the Ox Alpha explainer](https://cellcog.ai/blog/what-is-ox-alpha/), and on August 26 Z.ai confirmed to Bloomberg that it was GLM-5.3-Flash and released the weights, which we recorded in [the reveal post](https://cellcog.ai/blog/glm-5-3-flash/). Timing made the guess tempting: Ox Alpha landed inside Musk's "3 to 4 weeks" window. Timing is not evidence. As of September 8 no anonymous checkpoint on LMArena, OpenRouter, Yupp or Artificial Analysis has been attributed to Grok 4.7 by anyone with output to show.

## What 4.7 would land on top of

Grok 4.6 is the reference point, and xAI's [launch post](https://x.ai/news/grok-4-6) is unusually specific. Pricing "starts at $2 per million input tokens and $6 per million output tokens," with a fast variant at twice the price. It launched "available today in Cursor and Grok Build," with 2x included usage in both for the first week. xAI's own table puts it at 61 on the Artificial Analysis Intelligence Index, level with GPT-5.6 Sol Max and one point behind Fable 5 Max at 62, and at 65.9% on DeepSWE v1.1 against 70% for Fable 5 Max and 73% for Sol Max. Those are vendor numbers, labeled as such. Our record of the Grok Bot side is in [What is Grok Bot](https://cellcog.ai/blog/what-is-grok-bot/) and [Grok Bot pricing](https://cellcog.ai/blog/grok-bot-pricing/).

The 4.x line has been shipping fast, and the interval is what makes "a few weeks later" plausible at all.

*Table: Grok 4.x releases and the gap since the previous numbered release*

| Model | Announced by xAI | Days since previous |
|---|---|---|
| Grok 4 | July 9, 2025 | n/a |
| Grok 4.1 | November 17, 2025 | 131 |
| Grok 4.20 (public beta) | February 17, 2026 | 92 |
| Grok 4.5 | July 16, 2026 | 149 |
| Grok 4.6 | August 12, 2026 | 27 |
| Grok 4.7 | Musk target about September 11 | 30 if it holds |

Grok 4 Fast (September 19, 2025) and Grok 4.1 Fast (November 19, 2025) are left out of the table because they are speed variants of existing models, not new bases. The 4.20 row uses Musk's February 17 public-beta announcement; xAI's release notes date the API release March 10, and the 4.5 row uses the x.ai news date of July 16 where the release notes list API availability on July 8. Both readings are noted so nobody has to guess which one we used.

## Where Grok 4.7 will show up first

If 4.6 is the template, the order is API and Grok Build on day one, Cursor the same day, and the consumer surfaces after. Grok Bot is the open question. xAI's documentation names Grok 4.6 as the default model of Grok Build on the API and CLI. The [Grok Bot documentation](https://docs.x.ai/grok-bot/bots) and the [September 3 design post](https://x.ai/news/designing-grok-bot) describe persistence, memory, tools and the Bot roster without naming the model underneath. A model swap inside a persistent agent product is a bigger event than a new API id, because the agents that are already running change behavior under existing work, and xAI has not said how it handles that. When it does, that goes in the tracker.

## What we are watching for

We run production work on three model families and watch a fourth, so this list is an operator's.

- **A model id.** The moment docs.x.ai lists a grok-4.7 identifier or the release notes carry the line, this stops being a rumor page.
- **A launch post on x.ai/news.** Grok 4.6's arrived the same day as the API. We check the news feed and the docs several times a day.
- **The September 11 date.** Musk's "10 days" from September 1. A miss becomes a row in the scorecard, not a scandal; a hit is the first Grok date he has landed on the day.
- **Pricing relative to 4.6.** He said 4.7 is "slightly slower to serve." If the $2 and $6 rates move, the 4.6 pricing story on our [Grok Bot pricing](https://cellcog.ai/blog/grok-bot-pricing/) page changes with it.
- **The first published table.** Whether it leads with engineering and agentic rows, as the SpaceX-data claim implies, and whether any independent evaluator has a number on launch day.
- **Grok Bot.** Whether xAI names the model under Grok Bot for the first time, and whether running Bots move to 4.7 automatically.

## The tracker

This page updates when facts change, not when posts get louder. As of September 8, 2026: Grok 4.7 has been described by its founder, never by its company. No date, no spec, no benchmark, no price exist in any xAI document. The most likely picture is a real, larger model arriving somewhat after the September 11 target, the way 4.5 and 4.6 did. When xAI moves, the confirmed facts and a graded scorecard of every claim above land here the same day.

## Update log

This is a living page; when the story moves, the update lands here.

As of September 8, 2026: page opened. Latest Musk target is about September 11; no official movement to log.

## FAQ

**Has xAI announced a Grok 4.7 release date?**

No. Every date comes from Elon Musk's personal account on X. On July 24, 2026 he wrote 'Grok 4.6 in 2 weeks and Grok 4.7 in 4 weeks'; on July 28, that 4.7 would come 'a few weeks' after 4.6; on August 12, the day 4.6 launched, that it 'should be ready in 3 to 4 weeks'; and on September 1, 'Grok 4.7 comes out in 10 days'. xAI's own channels, the x.ai news page, the model directory and the API release notes, have not mentioned the model as of September 8.

**What are the Grok 4.7 specs?**

There is no published spec. Musk described it on July 28 as 'the 2.1T model', larger than the 1.5T Grok 4.6, 'better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency'. On July 21 he said SpaceX's engineering data, minus ITAR-restricted material, would be added 'during supplemental training of the 2T run', and on August 12 that 'initial training is complete' and the SpaceX data was being added. Context window, price, API name and benchmark scores are all unknown. The 4 to 6 trillion parameter figure that circulates traces to a July 24 NextBigFuture post that labels it speculation.

**Are the leaked Grok 4.7 benchmarks real?**

We could not find a real one. No sheet from xAI, LMArena, Artificial Analysis or any recognized evaluator names a tested model as Grok 4.7. Posts titled 'Grok 4.7 leaks' from late July carry no files and no source. The Grok scores that do circulate as leaks are from July 2025 and concern Grok 4. Musk's claims that 4.7 'will exceed all current models' (August 12) and 'has a good chance of exceeding all current models in intelligence' (August 15) are predictions, not results.

**Was Ox Alpha secretly Grok 4.7?**

No. Ox Alpha appeared as an anonymous stealth model on OpenRouter on August 20, 2026, and within a day people on X were guessing it was Grok 4.7 or Gemini 3.8 Flash. On August 26 Z.ai confirmed it was GLM-5.3-Flash and published the weights. We tracked that window as it happened in our Ox Alpha explainer and the GLM-5.3-Flash reveal post. As of September 8 no anonymous checkpoint on any leaderboard has been credibly attributed to Grok 4.7.

**Will Grok 4.7 be in Grok Bot?**

Probably eventually, but nothing official says so. Grok 4.7 is described as a base model, the successor to Grok 4.6. xAI's docs currently name Grok 4.6 as the default model of Grok Build, and Grok 4.6 shipped into Cursor and Grok Build on launch day, so that is the likely path for 4.7 too. The Grok Bot docs and xAI's September 3 'Designing Grok Bot' post do not name the model underneath Grok Bot, so we will log that when xAI does.

**Should I wait for Grok 4.7?**

Only if your work is blocked on something 4.6 cannot do. If you are choosing a model API, Grok 4.6 is live at $2 per million input tokens and $6 per million output tokens and 4.7 will arrive as a new model id when it ships. If you are choosing where your agents run, the model version matters less than the layer around it: memory that survives the conversation, a task board, an inbox, approvals. CellCog does not route to Grok today; our tiers run on Gemini 3.8 Flash, Claude Fable 5.1 and Opus 5, and a standing AI employee keeps its role across every engine swap. Plans start at $8 a month; you pay for the work, not the hire, and the cost depends purely on how much work you assign.

## Related

- [Grok Bot: Always-On AI Teammates From $20 a Month, All Sharing One Computer](https://cellcog.ai/blog/what-is-grok-bot/index.md)
- [Grok Bot Pricing Explained: What It Actually Costs](https://cellcog.ai/blog/grok-bot-pricing/index.md)
- [Gemini 4: Release Date, Leaks, and What Google Has Actually Said](https://cellcog.ai/blog/gemini-4-release-date/index.md)
- [GPT-6 Astra Is Out: Price, Specs, Benchmarks, Rollout, and What Is Still Open](https://cellcog.ai/blog/openai-astra-release-date/index.md)

---

Markdown alternate of https://cellcog.ai/blog/grok-4-7-release-date/. Try CellCog free, no credit card needed: https://cellcog.ai/signup