# What Is Grok 4.7, and Should Professionals Switch from 4.6? (September 2026)
> Grok 4.7 launched September 21, 2026 — same API price as 4.6, 40% more parameters, and a new fast variant in Cursor. Independent benchmarks show a widening gap to Claude and GPT-6. Here is what changed, what didn't, and whether it moves the needle for professionals.
**Author:** [Alex Lowe](https://theaicareerlab.com/about) — Founder, The AI Career Lab
**Published:** 2026-09-06
**Last reviewed:** 2026-10-13
**Canonical URL:** https://theaicareerlab.com/blog/grok-4-6-for-professionals-2026
**Category:** guide
**Tags:** Grok, Grok 4.7, xAI, SpaceXAI, AI tools, model update, AI at work, data privacy, professionals, 2026
---> _**Update (September 2026):** Grok 4.7 launched September 21, 2026 — same API price as 4.6, 40% more parameters, and a new Grok 4.7 Fast variant in Cursor and Grok Build. Independent benchmarks (Artificial Analysis Intelligence Index: 46, vs Claude Fable 5.1 and GPT-6 at 53) show the gap to the frontier widened. The rest of this guide has been updated throughout; the data privacy section below applies to 4.7 unchanged._

---

The straight answer on Grok 4.7: it is a free upgrade at the same API price, and for teams already using Grok 4.6 for coding through Cursor or the API, it is worth switching to. It is still not the right tool for writing, analysis, or anything a client reads. And the data defaults in the consumer app have not changed — that section still applies.

The rest of this guide explains who makes it, what 4.7 added, what it costs, and whether client data belongs in it.

## Who makes Grok now

Grok is built by xAI (now SpaceXAI). SpaceX acquired xAI on February 2, 2026, and press reports say the unit was rebranded SpaceXAI in early July. For professionals this mostly changes who you are contracting with: the entity on any API agreement or enterprise contract is now a SpaceX subsidiary.

The bigger structural change is Cursor. SpaceX closed its $60 billion acquisition of Cursor, the AI code editor, on August 14, 2026. Cursor is now a first-party Grok surface, which is why Grok's release notes emphasize Cursor performance. If your developers use Cursor, they are using a SpaceXAI product whether or not they ever open grok.com.

One more piece of context, attributed to press coverage rather than the company: TechCrunch and Gizmodo have reported that Musk told staff Grok is "losing the AI race to Anthropic," and that all of xAI's original co-founders have departed. Read the product on its merits, but know that the organization behind it is in flux.

## Which Grok is current

**Grok 4.7 is the newest model**, released September 21, 2026. It replaced Grok 4.6 (August 12, 2026) as the flagship. There is no Grok 5 — any page describing Grok 5 capabilities is describing something that does not exist.

The older models remain available: Grok 4.6 (`grok-4.6`), Grok 4.5 (`grok-4.5`), and the cheaper mid-tier Grok 4.3 (`grok-4.3`).

In the chat app, the picture is muddier. Paid tiers run the current flagship or the previous version depending on rollout timing, and the Free tier runs an older, lighter model. xAI has not published a clear table of which tier gets which model by default, so if the version matters to you, check the model picker after you sign in.

## What Grok 4.7 adds over 4.6

Three things are new in 4.7:

**More parameters.** Elon Musk put the parameter count at 2.1 trillion, roughly 40% more than Grok 4.6's 1.5 trillion. xAI describes this release as "built on a larger base model with extended reinforcement learning and improved self-verification mechanisms."

**A fast variant.** Grok 4.7 Fast is available in Cursor and Grok Build. No published price difference from 4.7 standard, but it is aimed at workflows that need speed over deep reasoning.

**GitHub Copilot availability.** GitHub is rolling out Grok 4.7 to all Copilot plans — Pro, Pro+, Max, Business, and Enterprise — in a phased launch. If your team uses GitHub Copilot, you may already have access.

**What did not change:** the 500,000-token context window, the API price structure, and the knowledge cutoff (May 2026 for 4.7).

## What the benchmarks say

xAI says 4.7 is its most capable model yet. Independent benchmarks draw a more qualified picture.

The Artificial Analysis Intelligence Index puts Grok 4.7 at **46**, versus Claude Fable 5.1 and GPT-6 at **53**. On Terminal-Bench 4.0, which tests agentic coding tasks, Grok 4.7 scores **26%** — behind GPT-6 Astra at 60% and Claude Fable 5.1 at 55%.

In other words: Grok 4.7 is a bigger model at the same price, but the gap to the top frontier models is wider, not smaller, than it was with 4.6. Whether that matters for your use case depends on whether you are comparing Grok to itself or to the current frontier.

For teams using Grok as a cheap coding tool — not as a replacement for the best available model — 4.7 is still a solid value at $2/$6 per million tokens. The Fast variant in Cursor may be the most practically useful change.

## What it costs

API pricing, per million tokens, from SpaceXAI's developer documentation (verified September 21, 2026):

| Model | Input | Cached input | Output | Context |
|---|---|---|---|---|
| Grok 4.7 | $2.00 | $0.50 | $6.00 | 500K |
| Grok 4.6 | $2.00 | $0.50 | $6.00 | 500K |
| Grok 4.5 | $2.00 | $0.30 | $6.00 | 500K |
| Grok 4.3 | $1.25 | $0.20 | $2.50 | 1M |

All prices double for requests above 200,000 tokens of context. Grok 4.7 and 4.6 are priced identically at this level.

Consumer plans changed on October 12, 2026. SpaceXSI now sells one plan family, **The X Subscription**, that covers Grok, Grok Bot, Cursor, and X Premium; its pricing page lists:

- Free: $0, limited Grok and a starter allowance for Cursor's coding agents
- Plus: $8 per month (Grok Chat, Voice, Imagine, Build, the Grok API, and Cursor with Grok models; no Grok Bot)
- Premium: $30 per month, 4x Plus usage, the latest Grok models and Grok Bot (SpaceXSI describes it as "the benefits of SuperGrok, Cursor Pro, and X Premium+")
- Super: $100 per month, 15x Plus usage
- Ultra: $200 per month, 50x Plus usage

Annual billing is cheaper (Premium is $300 a year). The older X Premium, SuperGrok, SuperGrok Heavy, and Cursor plans are closed to new sign-ups; existing subscribers can keep them or switch for a prorated refund. Check [x.ai/pricing](https://x.ai/pricing) before you buy.

The newest Grok models are a paid feature. If you are on the free tier, you may not be testing Grok 4.7.

## What it isn't good at

Writing, and any task where judgment and tone carry the value. This did not change with 4.7.

The early community verdict on Grok 4.6 was consistent: capable model for code, poor model for writing and judgment-heavy tasks. The Terminal-Bench 4.0 score for 4.7 (26%) suggests coding performance is still well behind the frontier despite more parameters. For client-facing documents, analysis, or anything a client will read, stay with Claude, ChatGPT, or Gemini.

For which model suits which kind of work, see our comparison of [ChatGPT, Claude, and Gemini for professionals](/blog/chatgpt-vs-claude-vs-gemini-for-professionals).

## Can you put work data in it?

This is the section to read twice. Grok has two front doors, and their data defaults are opposites. This has not changed with 4.7.

**The API and tools built on it.** xAI's security documentation says it never trains on API inputs or outputs without permission. Data is retained for 30 days, encrypted, and there is a zero-data-retention toggle in the developer console. xAI reports SOC 2 Type 2 certification. Those are the same commitments the major providers make for their business tiers, and they cover Cursor, Grok Build, and the other integrations that call the API. GitHub Copilot's Grok 4.7 integration goes through the API and carries GitHub's own data terms.

**The consumer grok.com app.** According to press coverage and the app's own settings, Grok trains on your chats by default unless you turn off "Improve the model" under Settings > Data, or use Private Chat for a given conversation. For users outside the EU, public posts on X are also used for training. Ireland's Data Protection Commission has an open investigation into xAI's use of European users' data. Our general guide to [whether AI trains on your data](/blog/does-ai-train-on-your-data) covers the same setting for every major provider; Grok is the one where the default is most clearly against you.

Then there are the incidents. These are the reasons compliance teams hesitate, and a work audience should know them:

- **July 12 to 14:** the Grok Build command-line tool uploaded users' entire Git repositories, including committed secrets, to a Google Cloud storage bucket, even when users had told it not to and had the opt-out enabled. xAI disabled uploads. Axios and The Next Web reported it; no public postmortem has been found.
- **August 19 to 24:** Grok returned gibberish output to users for several days, as reported by TechCrunch.
- **Late August:** a class action was filed alleging that Grok was trained on child sexual abuse material, reported by Gizmodo. This is an allegation, not a finding.
- **First half of 2026:** a Resemble AI report tied Grok to 87% of the deepfake files it could trace to a source model.

None of that changes the API's contractual terms. It does change how much benefit of the doubt a security reviewer is likely to extend, and it is fair to expect a longer vendor review for Grok than for its competitors.

## Grok Bot and the "AI teammates" pitch

On August 11, xAI launched Grok Bot in beta, positioned as "AI teammates" that run on their own cloud computer, sign into your apps, and carry out tasks while you do something else. It launched on SuperGrok, SuperGrok Plus, SuperGrok Heavy, and Cursor Pro+; since October 12, 2026, new customers get it on The X Subscription's Premium ($30/month) and higher plans. On September 3, xAI followed with Grok Bot for Enterprise, which adds access controls, network restrictions, and audit logging. No pricing has been published for the enterprise version.

The pitch is the same one every provider is now making: an agent with its own browser and credentials that acts on your behalf. Given the July incident above, where a Grok tool ignored an explicit opt-out, the sensible posture is to try Grok Bot on accounts and data you could afford to lose before connecting it to anything else.

Full guide: [What Is Grok Bot, and Should It Sign Into Your Work Apps?](/blog/what-is-grok-bot-ai-teammates).

## Bottom line for professionals

- **If you already use Grok 4.6 through Cursor or the API,** switch to Grok 4.7 — it's the same price and xAI calls it more capable. Try the Fast variant if speed is your constraint.
- **If your team is evaluating Grok for the first time,** the core verdict is unchanged: cheap, good enough for coding, behind the frontier on independent benchmarks, not right for writing or judgment-heavy work.
- **If you write, advise, analyze, or produce anything a client reads,** stay with Claude, ChatGPT, or Gemini. The benchmark gap to the top models widened with this release.
- **If you use the grok.com app at all,** turn off "Improve the model" in Settings > Data today, or use Private Chat. If your work is regulated or confidential, do not put it in the consumer app under any setting.
- **If you use GitHub Copilot,** Grok 4.7 is rolling out to your plan — check Copilot settings to see if it's available.

## Sources

1. [Models and pricing — SpaceXAI developer documentation](https://docs.x.ai/developers/models) *(verified September 21, 2026)*
2. [Release notes — SpaceXAI developer documentation](https://docs.x.ai/developers/release-notes)
3. [xAI launches Grok 4.7 at bargain prices, but benchmarks reveal a wide gap to Claude and GPT-6 — The Decoder, September 21, 2026](https://the-decoder.com/xai-launches-grok-4-7-at-bargain-prices-but-benchmarks-reveal-a-wide-gap-to-claude-and-gpt-6/)
4. [Grok 4.6 — xAI news, August 12, 2026](https://x.ai/news/grok-4-6)
5. [Security FAQ — SpaceXAI developer documentation](https://docs.x.ai/developers/faq/security)
6. [Introducing Grok Bot — xAI news, August 11, 2026](https://x.ai/news/introducing-grok-bot)
7. [Grok Bot for Enterprise — xAI news, September 3, 2026](https://x.ai/news/grok-bot-for-enterprise)
8. [SpaceXAI's Grok uploaded customer code to the cloud despite opt-outs — Axios, July 14, 2026](https://www.axios.com/2026/07/14/spacexai-grok-customer-data)
9. [Grok keeps sending gibberish responses to users — TechCrunch, August 20, 2026](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/)
10. [@JayaNayak21 on X, relaying Musk: "Grok 4.7 will be released in 10 days"](https://x.com/JayaNayak21/status/2096628005418451351)
## Frequently asked questions

### What is Grok 4.7?

Grok 4.7 is SpaceXAI's newest model, released September 21, 2026. It has roughly 2.1 trillion parameters — about 40% more than Grok 4.6's 1.5 trillion — and a 500,000-token context window. API pricing is the same as 4.6: $2 per million input tokens and $6 per million output tokens for requests under 200,000 tokens, doubling above that threshold. A faster variant, Grok 4.7 Fast, is available in Cursor and Grok Build. GitHub Copilot is rolling out Grok 4.7 in a phased launch across all plans.

### How is Grok 4.7 different from Grok 4.6?

The main differences are more parameters (2.1T vs 1.5T), a new Grok 4.7 Fast variant in Cursor and Grok Build, and formal availability in GitHub Copilot. API pricing and context window are unchanged. On independent benchmarks, Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index versus 53 for Claude Fable 5.1 and GPT-6, and 26% on Terminal-Bench 4.0's agentic coding test versus 55–60% for the frontier models. The benchmark gap to the top tier is wider with 4.7 than it was with 4.6.

### Is there a Grok 5?

No. As of September 21, 2026, the newest released Grok model is Grok 4.7, which shipped today. Grok 5 has not been announced with a release date.

### Who makes Grok now?

Grok is made by xAI, which SpaceX acquired on February 2, 2026. The unit was rebranded SpaceXAI in July according to press reports. SpaceX also closed a $60 billion acquisition of Cursor, the AI code editor, on August 14, 2026, so Cursor is now a first-party surface for Grok models.

### What is Grok 4.7 actually good at?

xAI positions Grok 4.7 as its best model for coding and knowledge work, and the added parameters and xhigh reasoning effort tier may help on complex multi-step tasks. That said, independent coding benchmarks (Terminal-Bench 4.0: 26%) put it well behind GPT-6 Astra (60%) and Claude Fable 5.1 (55%). If you are already using Grok 4.6 for coding via Cursor or the API, 4.7 is a free upgrade worth trying — particularly the Fast variant.

### What is Grok 4.7 bad at?

Writing and judgment-heavy work — the same weakness as 4.6. The benchmark gap to Claude and GPT-6 widened, not closed, on this release. For client-facing documents, analysis, or anything where tone matters, it is not the right choice.

### How much does Grok 4.7 cost?

Through the API, Grok 4.7 costs $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens for requests under 200,000 tokens of context. Prices double above that threshold. This is identical to Grok 4.6 pricing. Check grok.com for current consumer-tier pricing, as subscription tiers change frequently.

### Can I put client or work data into Grok 4.7?

The same answer as for 4.6: the API and tools built on it (Cursor, Grok Build) do not use your inputs for training without permission, keep data for 30 days encrypted, and offer a zero-data-retention setting. The consumer grok.com app trains on your chats by default unless you turn off 'Improve the model' under Settings > Data or use Private Chat. For regulated or confidential work, use the API path or do not use Grok.

---

*Canonical version: https://theaicareerlab.com/blog/grok-4-6-for-professionals-2026*
*This document is the Markdown companion served for AI crawlers and answer engines. See the canonical URL for the rendered version with navigation, related content, and interactive elements.*