Grok 4.7 Release Date: Out September 21, 2026, What's New
grok-4.7, a 500,000-token context window, and the same token rates as Grok 4.6. EvoLink added a live Grok 4.7 route on September 22, 2026.The direct answer
| Question | Confirmed answer (checked September 22, 2026) |
|---|---|
| When did Grok 4.7 come out? | September 21, 2026 (xAI announcement and developer docs) |
| Is Grok 4.7 out? | Yes, on the xAI API, Cursor, Grok Build, and model routers |
| What is the API model ID? | grok-4.7. The URL form grok-4-7 is not a valid request value |
| Is it available on EvoLink? | Yes, since September 22, 2026, with live pricing on the product page |
| Did the price change from Grok 4.6? | No. $2 input, $0.50 cached input, $6 output per million tokens, same as 4.6 |
| Is there a faster version? | Yes, but not on the API. Grok 4.7 Fast runs at twice the output speed and twice the price, and xAI offers it only in Cursor and Grok Build. EvoLink's route serves the standard model |
| What is the context window? | 500,000 tokens; prompts at or above 200,000 tokens bill every token at the long-context rate |
When did Grok 4.7 come out?
grok-4.7 went up at the same time.EvoLink activated its Grok 4.7 route on September 22, 2026. That is one day after the upstream launch and is a separate event: what EvoLink exposes, charges, and reports is an EvoLink fact, while the model's capabilities are xAI facts. If you searched for the release date because you are waiting for a specific channel, check that channel's own status rather than the announcement date.
What's new in Grok 4.7
The announcement describes Grok 4.7 as a larger base model with a longer reinforcement-learning run on harder, multi-hour tasks. xAI states the model is "better at verifying its own work and managing longer context," and positions it for coding, agent workflows, and knowledge work. These are the two claims most worth testing on your own workload.
xAI's published benchmark comparisons against Grok 4.6 are provider-reported numbers:
| Benchmark (xAI-reported) | Grok 4.6 | Grok 4.7 |
|---|---|---|
| Terminal-Bench 4.0 | 20.3% | 38.0% |
| CursorBench 4.0 | 40.4% | 46.3% |
| DeepSWE v1.1 | 65.2% | 71.0% |
| EEBench | 53.0% | 64.0% |
| AA-Briefcase v1.1 (Elo) | 1,546 | 1,657 |
| GDPval (Elo) | 1,605 | 1,695 |
| Harvey legal agent benchmark | 15.8% | 19.6% |
| HealthBench Professional | 48.5% | 56.7% |
xhigh setting uses about 81,000 output tokens per index task, against 36,000 for Grok 4.6 at high. The Grok 4.7 vs 4.6 guide puts both sets of numbers side by side and explains what the token difference does to your bill.low, medium, high (default), and xhigh. Image input means the model can read visual material; it is not an image-generation model.Grok 4.7 Fast
xAI also serves a Fast variant "with twice the output speed at twice the price." It is the same model on faster infrastructure, and xAI's pricing page states it is available only through Cursor and Grok Build: it is not on the public xAI API, and Grok Build's free tier does not include it. Within those products it costs $4 input, $1 cached input, and $12 output per million tokens below 200,000 prompt tokens, and $6, $1.50, and $18 above.
grok-4.7. The product page shows which model is routed and its current rates.Upstream availability and EvoLink availability are different facts
grok-4.7. Cached input, reasoning effort, and xAI's server-side tools follow the configuration shown in the product page's API section.Before it was released
The pre-release statements explain why the launch slipped and what the model was meant to fix. They are kept here as dated history, not as current status.
- Early September 2026: Musk said Grok 4.7 would arrive in about ten days (source).
- Mid-September 2026: Musk said the model needed more time, citing stopping difficult tasks too early and not checking its own work (source). The release notes now claim improvements in exactly those areas.
- Mid-September 2026: Musk placed the intended model roughly around Claude Opus 5.0, with differences between tasks (source). The Grok 4.7 vs Claude Opus 5 guide checks that claim against third-party numbers.
What to verify before production
A successful first response starts an evaluation; it does not finish one.
- Confirm the model ID in the response. Send
grok-4.7and check that the returned model field matches. - Check usage against the price tiers. Look at input, cached input, output, and reasoning usage on a representative request. Prompts at or above 200,000 tokens move the whole request to the long-context tier.
- Retest tools and structured outputs. Function arguments, JSON Schema validity, streaming completion events, and tool-failure recovery are where a version change shows up first.
- Record cost per accepted task. Divide all billed attempts, including retries and tool charges, by the tasks your acceptance rule passes. Given the third-party token-use numbers above, this is the check that decides whether the same rate means the same bill.
Update record
FAQ
When did Grok 4.7 come out?
September 21, 2026. xAI published the announcement and the developer model page on that date. EvoLink's route followed on September 22, 2026.
Is Grok 4.7 out now?
grok-4.7.What is the Grok 4.7 status on EvoLink?
Why was Grok 4.7 delayed?
Before launch, Musk said the model needed more work on finishing difficult tasks and checking its own results. The release notes now claim improvements in those areas. Verify them on your own tasks.
Is Grok 4.7 more expensive than Grok 4.6?
xhigh uses more than twice the output tokens of Grok 4.6 at high per task, so total cost for the same work can differ. The upgrade guide covers how to measure it.What is Grok 4.7 Fast?
The same Grok 4.7 model served on faster infrastructure at twice the token price: $4 input and $12 output per million tokens below 200,000 prompt tokens. xAI offers it only in Cursor and Grok Build, not on the public API. EvoLink's route serves the standard model.
Does Grok 4.7 generate images?
No. The documented modalities are text and image input with text output.
What changed in X Search billing?
From September 21, 2026 at 12:00 PT, xAI bills X Search per fetched post and per fetched user profile rather than per call, for Grok 4.5, 4.6, and 4.7.
