> ## Content Index
> Fetch the complete content index at: https://www.edgewisely.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Grok 5: Release Date, Specs & What We Know
- URL: https://www.edgewisely.com/grok-5/
- Published: 2026-11-16T04:30:40.000Z
- Updated: 2026-11-16T04:30:40.000Z
- Description: Grok 5 isn't released as of Nov 2026. Here's what SpaceXAI confirmed vs what Musk claimed (6T params, AGI), plus the real Grok 4.7 you can use today.
- Author: John Karpentar
- Tags: AI, Opinion

Grok 5 is **SpaceXAI's** (formerly xAI's) next flagship model. As of November 2026 it is **not released**, has **no official specs** and **no confirmed date** — only Elon Musk's claims of roughly **6 trillion parameters** and near-AGI ability, none of them independently verified. Here's what's confirmed, what's rumored, and what to actually use today.

## TL;DR

- **Not released.** As of November 2026, there is no Grok 5\. No launch date has been officially confirmed by SpaceXAI.
- **No official specs.** The "6 trillion parameters" and "AGI" figures trace to Musk's statements, not to any spec sheet, model card, or benchmark you can check.
- **Musk's claims are unverified.** He called Grok 5 the future "smartest AI in the world" at a November 2025 investor conference and targeted an early-2026 launch. That window has passed.
- **The real model today is Grok 4.7**, released **September 21, 2026**: Artificial Analysis Intelligence Index **v5.0.1 = 35**, **$2.00/$6.00** per 1M tokens, **500k** context window.
- **Verdict:** Build on Grok 4.7, Claude Opus 5.5, or Gemini 4 now. Don't architect anything around Grok 5 promises.

## Is Grok 5 out yet?

No. As of **November 16, 2026**, Grok 5 has not shipped. There is no public API, no model card, and no pricing.

The most recent released model is **Grok 4.7** (September 21, 2026). Neither Wikipedia's [Grok (chatbot)](https://en.wikipedia.org/wiki/Grok%5F%28chatbot%29?ref=edgewisely.com) page nor the [SpaceXAI](https://en.wikipedia.org/wiki/SpaceXAI?ref=edgewisely.com) page documents a Grok 5 release, specs, or date. Every "Grok 5" page ranking in search today is speculation — Reddit threads, YouTube videos, and "what we know so far" blogs working from the same handful of Musk posts.

Here's the distinction that matters: a model being *in training* is not a model being *released*. Musk said Grok 5 was in training back in November 2025\. Training runs produce checkpoints, not products. Until SpaceXAI publishes a model card and opens API access, nothing about Grok 5 is a usable fact.

## What has Musk actually confirmed about Grok 5?

Officially, SpaceXAI has confirmed nothing. There is no company spec sheet or launch announcement. What exists is a set of **dated claims from Musk**, mostly on X and at investor events — and those are claims, not confirmations.

The clearest marker: at the Baron Capital investment conference in **mid-November 2025**, Musk said Grok 5 was "in training" and would be "the smartest AI in the world by a significant margin," [as reported by IBTimes](https://www.ibtimes.co.uk/elon-musk-reveals-grok-5-coming-2026-jaw-dropping-6-trillion-parameters-1755754?ref=edgewisely.com). He tied it to a target launch in **early 2026**.

That target slipped. It is now late November 2026, four shipped Grok releases later (**4.5**, **4.6**, **4.7**, plus point updates), and Grok 5 still isn't out. This matters for how you weight any new date: Musk's model-timing estimates have a documented history of slipping, and the early-2026 Grok 5 target is now a concrete example.

So treat every Grok 5 figure below as **Musk's claim or third-party speculation** — never as a confirmed spec.

## Grok 5 rumored specs: 6 trillion parameters and AGI?

The headline numbers are unconfirmed. The two that dominate search — **\~6 trillion parameters** and **AGI** — both come from Musk's November 2025 remarks, not from any benchmark or model card.

On size: Musk reportedly described Grok 5 as roughly **6 trillion parameters**, framed as about double the size of Grok 3 and Grok 4\. No one outside SpaceXAI has verified this, and parameter count alone doesn't determine capability — data quality, training compute, and post-training matter as much or more.

On AGI: this is where the hype runs hottest and the evidence runs thinnest. Musk suggested higher "intelligence density" could approach human-level intelligence and put the odds at roughly **10%** by his own estimate. A 10% chance, stated by the person selling the model, is not a confirmed capability. "AGI" also has no agreed benchmark, so the claim isn't even falsifiable in the way an Intelligence Index score is.

On compute: Grok 5 is widely reported to be training on **Colossus**, SpaceXAI's Memphis supercomputer. On **December 30, 2025**, the company announced plans to scale toward **at least 1 million GPUs and nearly 2 gigawatts**, [per Wikipedia's SpaceXAI page](https://en.wikipedia.org/wiki/SpaceXAI?ref=edgewisely.com). That's a real, documented infrastructure plan. Whether it has already converted into a frontier-class released model is exactly the open question.

One naming note so you're not confused by sources: SpaceX acquired xAI in an **all-stock deal announced February 2, 2026**; the unit rebranded to **SpaceXAI in July 2026**, and shifted further toward "SpaceX Super Intelligence" (handle **@SpaceXSI**) in October 2026, though the legal name remains SpaceXAI LLC. Same maker, three names in a year.

## What can you use right now? Grok 4.7 in context

Use **Grok 4.7**. It's the real, shipping model, released **September 21, 2026**, and it's what every Grok 5 rumor is implicitly measured against.

Here are the verified figures from [Artificial Analysis's Grok 4.7 page](https://artificialanalysis.ai/models/grok-4-7?ref=edgewisely.com), as of November 2026:

- **Intelligence Index (v5.0.1): 35** at xhigh and high reasoning effort; **31** at low effort. That's **below the comparison median of 41** on the same index.
- **Price: $2.00 input / $6.00 output** per 1M tokens, with a **75% cache discount** (blended \~**$1.35**/1M at a 7:2:1 cache/input/output ratio). We broke the full cost picture down in our [Grok API pricing guide](https://www.edgewisely.com/grok-api-pricing/).
- **Context window: 500k tokens.**
- **Output speed: \~66 tokens/sec** at xhigh.

Read that honestly. Grok 4.7 is **priced competitively** and has a **large context window**, but on independent intelligence scoring it sits **below the median** of the models it's compared against — around 15th on Artificial Analysis's ranking, not near the frontier.

The lineage shows the cadence SpaceXAI is actually shipping, [per Wikipedia](https://en.wikipedia.org/wiki/Grok%5F%28chatbot%29?ref=edgewisely.com):

| Model     | Released     | Notes                                                                                 |
| --------- | ------------ | ------------------------------------------------------------------------------------- |
| Grok 4.20 | Feb 2026     | —                                                                                     |
| Grok 4.3  | Apr 2026     | —                                                                                     |
| Grok 4.5  | Jul 8, 2026  | $2/$6; Musk called it "Opus-class"; trained on tens of thousands of Nvidia GB300 GPUs |
| Grok 4.6  | Aug 12, 2026 | —                                                                                     |
| Grok 4.7  | Sep 21, 2026 | Trained partly on Cursor data; AA Index v5.0.1 = 35                                   |

That's five releases in roughly seven months — fast point iteration, but all in the 4.x line. Grok 5 would be the first major-version jump since Grok 4.

## Grok 5 vs the current frontier: the bar it has to clear

Here's the comparison that actually matters — shipping reality versus Musk's claims:

|                                    | Grok 4.7 (shipping)     | Grok 5 (Musk's claims)                              |
| ---------------------------------- | ----------------------- | --------------------------------------------------- |
| **Status**                         | Released                | **Rumored / unconfirmed**                           |
| **Release date**                   | Sep 21, 2026            | **Unconfirmed; early-2026 target slipped**          |
| **Parameters**                     | Not disclosed           | **\~6T (Musk, \~Nov 2025) — unverified**            |
| **AA Intelligence Index (v5.0.1)** | **35** (median 41)      | **No score exists — not released**                  |
| **Price (per 1M)**                 | **$2.00 / $6.00**       | **Unconfirmed**                                     |
| **Context window**                 | **500k tokens**         | **Unconfirmed**                                     |
| **Training cluster**               | SpaceXAI infrastructure | **Colossus (reported) — attributed, not confirmed** |

Every Grok 4.7 cell is a sourced figure. Every Grok 5 cell is a claim or a blank. That asymmetry *is* the story.

Now the bar. On Artificial Analysis's current Intelligence Index, the top of the board sits far above Grok 4.7's 35 (as of November 2026):

| Model             | Intelligence Index |
| ----------------- | ------------------ |
| Claude Fable 5.2  | 65                 |
| Gemini 4 Carbon   | 60                 |
| Claude Opus 5.5   | 59                 |
| GPT-6 Astra       | 57                 |
| GPT-6.1 Sol       | 55                 |
| Gemini 4 Hydrogen | 54                 |
| **Grok 4.7**      | **35**             |

To be "the smartest AI in the world by a significant margin," Grok 5 would need to clear **\~65** — nearly double Grok 4.7's current score, and comfortably past Claude Fable 5.2 at the top. That's the real, measurable gap the hype has to cross. No shipped Grok model has been near it. For a sense of where today's leaders land, see our [Grok vs Gemini comparison](https://www.edgewisely.com/grok-vs-gemini/).

![Bar chart: Grok 4.7 scores 35 on the Artificial Analysis Intelligence Index while frontier models score 54 to 65](https://storage.ghost.io/c/54/5a/545a66b3-60ef-480c-80ae-765bac52f6ec/content/images/2026/11/grok_chart.png)

Chart: Edgewisely, data from [Artificial Analysis](https://artificialanalysis.ai/leaderboards/models?ref=edgewisely.com) (as of Nov 2026)

## Should you wait for Grok 5?

No. Build on what ships. There is no date, no spec, and no benchmark for Grok 5 — you cannot plan a roadmap around a model that doesn't exist.

- **If you need Grok specifically:** use **Grok 4.7** today. It's cheap and has a 500k context window; just don't expect frontier-level reasoning at its AA score of 35.
- **If you need frontier capability now:** use **Claude Opus 5.5**, **Gemini 4**, or **GPT-6** — all materially higher on the independent index.
- **Wait for Grok 5 only if** you have a non-urgent project *and* a specific reason to prefer SpaceXAI's stack, and even then, wait for the model card and an independent benchmark before committing.

## Our take

The most useful thing you can do with the Grok 5 hype cycle is ignore it and buy on shipped capability. That sounds dismissive. It's actually the disciplined position, and here's the reasoning.

Model-timing claims from SpaceXAI have a track record of slipping. Grok 5 was described as "in training" with an early-2026 launch target back in November 2025\. It is now late November 2026\. In the intervening year the company shipped Grok 4.5, 4.6, and 4.7 — real, useful point releases — but not the major-version model it had already named. That's not a scandal; fast-moving labs reprioritize constantly. But it tells you exactly how much weight to put on a stated date: enough to note it, not enough to plan around it. If you built a product assumption on "Grok 5, early 2026," you'd have been wrong for most of a year.

So the interesting question isn't the rumored 6 trillion parameters. Parameter count is the least informative number in modern AI. It's a claim that's easy to make, impossible to verify from outside, and weakly correlated with the thing you actually care about — how the model scores and behaves. The real question is whether SpaceXAI can convert its very real compute advantage into *index position it has not yet shown*. Colossus and the December 2025 plan toward a million GPUs and \~2 gigawatts are genuine, documented infrastructure. That's a serious bet on scale. But Grok 4.7 lands at an Artificial Analysis Intelligence Index of 35 — below the comparison median of 41, and far below a frontier that now sits in the high 50s to mid 60s. The gap between "most compute" and "best model" is precisely the gap SpaceXAI hasn't closed in its shipping products. (That's my read, not a company statement.)

What would a genuinely AGI-class Grok 5 need to do? Concretely, not rhetorically: clear the top of an independent leaderboard it doesn't control, hold up on tasks it wasn't tuned for, and do it at a price and latency you'd actually deploy. "Smartest AI in the world by a significant margin" is a testable claim the moment the model ships and someone runs it. Until then it's marketing, and the honest label is *unverified*. I'd apply the same skepticism to any lab — this isn't a SpaceXAI-specific standard. If SpaceXAI's data practices matter to you, we covered [how Grok handles your data](https://www.edgewisely.com/is-grok-safe-to-use/) separately.

The real risk here is building on unreleased-model promises. If you're a founder or an engineer, every hour you spend designing around Grok 5's rumored specs is an hour spent on fiction. The models on the board today — Opus 5.5, Gemini 4, GPT-6, and yes, Grok 4.7 for cheap, long-context work — are the ones you can benchmark, price, and ship against. They have model cards. They have numbers. They exist.

My verdict: **wait for Grok 5 only if** you have a genuinely non-urgent project and a specific reason to be on SpaceXAI's stack — and even then, wait for the model card and an independent benchmark, not the launch tweet. **Don't wait if** you have anything to ship this quarter; the frontier is already well ahead of where Grok 5 would start, and you can use it today. For most teams, Grok 5 should be a calendar reminder to re-benchmark, not a line in your architecture.

## Frequently Asked Questions

### When is Grok 5 coming out?

No confirmed date exists as of November 2026\. Musk targeted an early-2026 launch in statements around November 2025, but that window passed without a release. SpaceXAI has not published an official date, model card, or API. Treat any specific "release date" you see as speculation, not confirmation.

### Is Grok 5 released yet?

No. As of November 16, 2026, Grok 5 is not available. There is no public API, pricing, or spec sheet. The most recent shipped model is Grok 4.7, released September 21, 2026\. Neither Wikipedia's Grok nor SpaceXAI page documents a Grok 5 release.

### Will Grok 5 be AGI?

Unknown and unconfirmed. Musk suggested Grok 5 could approach human-level intelligence and estimated roughly a 10% chance, by his own account, at a November 2025 investor event. "AGI" has no agreed benchmark, and no independent test of Grok 5 exists because the model hasn't shipped. Treat AGI claims as marketing until proven.

### How many parameters does Grok 5 have?

Not officially disclosed. Musk reportedly described Grok 5 as roughly 6 trillion parameters in November 2025, framed as about double Grok 3 and Grok 4\. No one outside SpaceXAI has verified this, and parameter count alone is a weak predictor of real-world capability. Treat 6 trillion as an unconfirmed claim.

### What's the difference between Grok 4.7 and Grok 5?

Grok 4.7 is a real, shipping model (released September 2026): Intelligence Index v5.0.1 of 35, $2/$6 per 1M tokens, 500k context. Grok 5 is an unreleased next-generation model with no confirmed specs — only Musk's claims of \~6 trillion parameters and near-AGI ability. One you can use today; the other doesn't exist yet.

**Editor's note — sources:** Grok 4.7 figures (Intelligence Index v5.0.1 = 35, $2/$6 per 1M, 500k context, \~66 tok/s, released Sep 21 2026) are from [Artificial Analysis](https://artificialanalysis.ai/models/grok-4-7?ref=edgewisely.com) as of November 2026\. Release lineage and corporate history are from Wikipedia's [Grok (chatbot)](https://en.wikipedia.org/wiki/Grok%5F%28chatbot%29?ref=edgewisely.com) and [SpaceXAI](https://en.wikipedia.org/wiki/SpaceXAI?ref=edgewisely.com) pages. Musk's Grok 5 claims (\~6T parameters, "smartest AI in the world," \~10% AGI odds, early-2026 target) trace to his November 2025 Baron Capital remarks as reported by [IBTimes](https://www.ibtimes.co.uk/elon-musk-reveals-grok-5-coming-2026-jaw-dropping-6-trillion-parameters-1755754?ref=edgewisely.com); every Grok 5 figure is a claim, not a confirmed specification. Frontier Intelligence Index scores are from the [Artificial Analysis leaderboard](https://artificialanalysis.ai/leaderboards/models?ref=edgewisely.com). Analysis in "Our take" is Edgewisely's opinion, labeled as such.