xAI: Grok 4.7 keeps the 4.6 price, and opens a cyber door by invitation
On 21 September 2026, xAI announces Grok 4.7. The banner says a model twice as fast and half the price of comparable models. The body says it is served at the same price and speed as Grok 4.6. Those two sentences do not say the same thing.
Getting started
Original AI-generated illustration · SecuFocus
At a glance
Key points
Take the page seriously on what is priced: 2 dollars per million input tokens, 6 dollars output, and a fast variant at twice the output speed and twice the price. The model is in Cursor, Grok Build and the API. A question sent there leaves for xAI. Cyber red-team access is by invitation, not a public setting.
What changes
The announcement page is dated 21 September 2026. It presents Grok 4.7 as xAI’s most capable model for coding and knowledge work. The banner says: twice as fast, at half the price of comparable models. The next paragraph says something else: served at the same price and speed as Grok 4.6.
The announced starting price is 2 dollars per million input tokens and 6 dollars output. A fast variant doubles output speed and doubles the price. Grok 4.7 is announced in Cursor, Grok Build, the Grok API, third-party coding harnesses, routers and cloud platforms. This is not an application you install to keep the model on disk.
| Sentence on the page | What it compares |
|---|---|
| Twice as fast, half the price | Models called comparable, with no list in that sentence |
| Same price, same speed | Grok 4.6 |
| Fast variant | Twice the output speed, twice the price |
| Starting price | $2 input, $6 output, per million tokens |
Who it is for
The page aims at code and long work: documents, presentations, tasks that take hours. If you do not call the API and do not use Cursor or Grok Build, the announcement does not change today’s messaging app.
The passage that matters for a security decision is narrower. xAI says it has a new safeguard stack, and that it is giving selected partners invite-only access to red-team capabilities, for defensive research. That is not a switch in the public settings. It is a separate door.
What leaves
Using Grok 4.7 through the API, Cursor or a router means sending the context to the service that hosts the model. The page does not describe an open weight you can download. A low price does not bring the question back onto your machine.
On cyber, xAI says 3.3 percent of risky dual-use prompts get through on its HackerBench v0.3, and that the model rarely blocks legitimate security work. That is their measurement, on their bench. A refusal percentage is not a guarantee that your sensitive prompt will be stopped, or that an ordinary prompt will pass.
What we did not verify
We did not call the API, open Cursor, or rerun CursorBench, DeepSWE or HackerBench. The percentages on the page remain xAI’s claims. The table also compares other models. We do not carry those columns over as a ranking.
The page publishes an install command. Copying it here would not help the decision. A script downloaded and run at once is an act of trust toward x.ai, separate from the choice of model.
The decision
If you are looking for a local model, this announcement is not your path. If a coding tool already calls Grok, the useful change is the price and the place where the repository you let it read is sent.
Do not read the “half price” banner as the price against Grok 4.6. The page itself separates the two comparisons. The priced figure is 2 and 6 dollars per million tokens, plus a fast variant at double.
Check and explore
Sources for this article
Numbers connect each reference to the passages that use it. Dates show when the documentation was consulted.
This article draws on the sources above. The exercises are for you to try on your devices; SecuFocus does not present them as tests carried out by its editorial team. Interfaces and features can change. Method and corrections.
Cite this article
Keep this reference with the article when you save or share it.