1. Home
2. Guides
3. xAI

Grok 4.7, Grok 4.8 and Grok Build: which Grok is out and what each one is for

Grok 4.7 shipped ten days after Musk said it needed a few more days, and the next one has a size but no date.

By Think Facility · September 23, 2026 · xAI models

Grok 4.7 is out. It reached xAI's API on September 21, 2026, and Grok 4.8 isn't out yet. The latest public word on 4.8 that I could find from anyone at xAI is Elon Musk's post of September 14, saying it "will finish training this week and start RL" (RL is reinforcement learning, the second stage of training).

That leaves four Grok names people ask about, and they're easy to mix up. Here's where each stands, from xAI's own API release notes and price list.

Every recent Grok, by date

Dates from xAI's API release notes and prices from its pricing page, both read September 23, 2026. Prices are for prompts under 200K tokens.

One thing stood out when I lined the prices up: Grok 4.5, 4.6 and 4.7 all cost the same. xAI's Grok 4.7 announcement says it's "Served at the same price and speed as Grok 4.6", so an upgrade costs you nothing per token. The only reason to stay on an older one is a prompt you've tuned and don't want to retest.

Another thing you'll notice on xAI's pages is the company name. The release notes now call Grok 4.7 "SpaceXAI's frontier model", and that's the name in the site footer too.

Grok 4.7: what changed

xAI says Grok 4.7 "uses a new, larger base model" than 4.6, trained with a longer reinforcement learning run on tasks that take many hours. The API version reads 500K tokens of context, and its knowledge runs to May 2026.

The other change is aimed at xAI's own products. The launch post says Grok 4.7 was trained to understand the Grok Bot harness natively. Grok Bot went live on August 11, 2026, and xAI describes it as durable AI teammates working on a persistent cloud computer, with messaging, approvals, connectors and routines. xAI says the same training makes 4.7 better at conversation and general knowledge work.

It was held back once. On September 11, Musk wrote that it "needs a few more days to cook", and gave a reason I didn't expect to read from a CEO: the model still gave up on hard tasks too early and wasn't checking its work carefully enough. It came out ten days later, and checking its own work is one of the things the launch post leads on.

There's also a Fast version at twice the token rates. You can't call it on the public API. The release notes say it's only in Cursor and Grok Build.

Where you can run Grok 4.7

xAI launched it in Cursor and in its own Grok Build tool on day one, and on the API as grok-4.7. The announcement also lists third-party coding harnesses, model routers and cloud platforms, without naming which. On the API you pick how hard it thinks with a reasoning effort of low, medium, high or xhigh, and it defaults to high.

Two developer details matter before you switch. The models page says logprobs (the per-token probability scores some tools read) aren't supported from Grok 4.20 onward, and are silently ignored if you send them. And xAI says it has started giving select cybersecurity partners invite-only access to Grok 4.7's red-team capabilities, which is the same gated-access idea Anthropic uses for Mythos.

How it stacks up, by xAI's own numbers

xAI's launch table puts Grok 4.7 against Grok 4.6, GPT-5.6 Sol and Claude Fable 5.1. It wins some rows and loses others by a lot, and xAI printed the losses along with the wins.

All figures from xAI's Grok 4.7 announcement, September 21, 2026, including its numbers for Fable 5.1. Grok 4.7 at xhigh effort, Fable 5.1 at max.

Musk's own ranking, posted a week before the launch, was blunter than the table. He wrote that Grok 4.7 "should be roughly on par with Opus 5.0, not 5.1", and that Grok 4.9 is "probably Astra/Fable class". (There's no Claude Opus 5.1, as it happens. The 5.1 generation shipped as Fable, which we've sorted out here.)

What Grok Build is

Grok Build is two things with one name, and that's where most of the confusion comes from.

First, it's a coding tool. xAI put Grok Build into beta on May 14, 2026 as a terminal app you install with one command, and it runs interactively or headless in scripts. Grok 4.7 and its Fast version run inside it.

Second, there's a model called Grok Build 0.1 (grok-build-0.1), added on May 19 as an early-access release. It's a coding model, and it's the cheapest text model on xAI's price list at $1 in and $2 out, with a 256K window. If you've got old code calling grok-code-fast-1, that name is now an alias for it.

If you're weighing Grok Build 0.1 against Grok 4.6, xAI's models page has since settled it in a way that makes both answers stale: for code, and for everything else that isn't images, video or voice, it says to use Grok 4.7. Build 0.1 is still the cheapest per token, so it's worth a test run when cost is what you're fighting.

Watch the rate limits if you go that way. The model pages give Grok Build 0.1 37 requests a second and 10 million tokens a minute, against 150 requests and 50 million tokens for Grok 4.6. Neither one accepts Batch API jobs.

The long-context Groks

The newest Groks aren't the biggest readers. Grok 4.7, 4.6 and 4.5 take 500K tokens, while the older Grok 4.3 and Grok 4.20 take a million at $1.25 in and $2.50 out. Every one of them doubles its rates once a prompt reaches 200K tokens, and the whole request is billed at the higher rate, so a 250K prompt costs twice what a 190K one would per token. Sending requests through xAI's US regional endpoint adds 10% on top.

Grok 4.8, 4.9 and 5

What Musk has said about the next three, all on September 14, 2026: Grok 4.8 is "a 2.5T model trained with our new C++ software stack" and "will be a noticeable improvement". 2.5T is the model's size in parameters, trillions of them. In the ranking post he also wrote that xAI needs to fix Grok 4.7's multimodal performance, meaning how it handles images alongside text. Grok 4.9 comes after 4.8, and Grok 5 after that, of which he wrote "maybe better than anything. We shall see."

None of the three has a date, a model page or a price. xAI's release notes are where it'll show up first, since every Grok in the table above has an entry there.

The newest entry in those notes today is Grok 4.7, dated September 21, 2026.