xAI announced Grok 4.7 on September 21, 2026 for coding and knowledge work. For developers, the key is to separate where the model runs from what it costs: the direct xAI API lists grok-4.7 at $2 per million uncached input tokens, $0.50 per million cached input tokens and $6 per million output tokens below the long-context tier. A prompt reaching the roughly 200,000-token long-context boundary moves the whole request to higher rates; xAI’s pages differ at exactly 200,000. Grok 4.7 Fast is available through Cursor and Grok Build, but not the public xAI API.
Use the API table below for direct xAI calls, Cursor’s own plan and context rules inside Cursor, and the product’s current controls in Grok Build. xAI’s launch benchmarks are vendor-reported snapshots, not an independent or up-to-date ranking against every model available today.
What Grok 4.7 is
The xAI API model ID is grok-4.7. The current model page lists a 500,000-token context window, text and image input, text output, and reasoning effort levels of low, medium, high and xhigh, with high as the default. xAI describes 4.7 as its most capable model for coding, agentic tasks and knowledge work; that is the vendor’s characterization, not a result from an independent test in this guide. Check xAI’s current model page before building against a specific capability or limit.
Choose the access surface before comparing prices
Grok 4.7 is listed across the xAI API, Cursor and Grok Build. These are different products with different entitlement and billing rules. A model router may also add its own rate or show a third-party route price, so do not treat every listing as the xAI API rate.
| Surface | What the current docs establish | Access and billing boundary |
|---|---|---|
| xAI API | Model ID grok-4.7; 500K context; standard model only. | Use xAI’s global API rates below. Fast is not offered on the public API. The US regional endpoint costs 10% more. |
| Cursor | Every paid plan lists Grok 4.7. Cursor bills it through the Cursor Models usage pool. | Start (India) is fixed at medium effort and non-Fast. Pro or higher allows effort selection and Fast. Cursor has its own long-context threshold above 256K. |
| Grok Build | xAI’s launch page and Build changelog list Grok 4.7. | Fast is available, but the Grok Build free tier does not include it. Check Build’s current plan and usage controls for the session you intend to run. |
| Consumer Grok | The sources checked here establish API, Cursor and Build access. | They do not establish that a consumer subscription includes Grok 4.7. Do not infer app-plan access from API availability. |
Cursor’s listed standard rates are $2 input, $0.50 cached input and $6 output per million tokens; Fast is $4, $1 and $12. Cursor says prompts above 256K use twice standard rates, or three times standard rates in Fast, up to its 500K window. Those are Cursor-specific rates and limits, not the xAI API’s approximately 200K long-context boundary. Cursor’s Grok 4.7 plan guide also says enterprise administrators manage model access.
For Build access, controls and product behavior, see our Grok Build access guide. The Grok Bot’s persistent cloud-computer workflow is a separate product built around a model; it is not another name for the Grok 4.7 model itself.
Direct xAI API pricing, including the long-context tier
xAI quotes direct API text prices per million tokens. Its pricing table labels long context at 200,000 tokens or more, while model and release-note prose describes the higher tier as applying when a prompt exceeds 200,000. The docs therefore differ at exactly the boundary. The examples here use 150,000 and 220,000 prompt tokens, not exactly 200,000; check the live billing page before estimating a request at the boundary.
| Prompt size | Uncached input / cached input / output per 1M | Billing rule |
|---|---|---|
| Below 200K | $2.00 / $0.50 / $6.00 | Short-context rates apply. |
| Long context (pricing table: ≥200K) | $4.00 / $1.00 / $12.00 | Once the prompt reaches the long-context threshold, xAI says all tokens in that request use the long-context rates. |
For example, a 150,000-token prompt with 130,000 uncached tokens and 20,000 cached tokens, plus 3,000 output tokens, costs about $0.288 at the below-200K rates: (130,000 × $2 + 20,000 × $0.50 + 3,000 × $6) ÷ 1,000,000. A 220,000-token prompt with 200,000 uncached and 20,000 cached tokens, plus the same 3,000 output tokens, costs about $0.856 at long-context rates: (200,000 × $4 + 20,000 × $1 + 3,000 × $12) ÷ 1,000,000. These illustrations exclude server-side tool invocation charges and any regional premium.
The US regional endpoint, https://us.api.x.ai/v1, applies a 1.1× premium to token rates. xAI also lists separate charges for server-side tools such as web search, X search and code execution. If a workload uses those tools, add their invocation charges to token usage; the model table alone is not the request’s full cost. See xAI’s current API pricing.
Grok 4.7 Fast: where it is available and what the rate table says
xAI describes Fast as the same Grok 4.7 model on faster infrastructure. Its pricing page says Fast is available only through Cursor and Grok Build, not the public xAI API; the Build free tier excludes it. In Cursor, Fast is available on Pro or higher, not the fixed medium/non-Fast Start plan.
| xAI Fast prompt band | Input / cached input / output per 1M | Scope |
|---|---|---|
| Below 200K | $4.00 / $1.00 / $12.00 | Fast rate table; Cursor and Grok Build only. |
| Above 200K | $6.00 / $1.50 / $18.00 | Long-context Fast rates; not a public API option. |
There is a mismatch within xAI’s pricing page: its prose calls Fast twice the standard token rates, but the published long-context row is 1.5 times the standard long-context row ($6/$1.50/$18 versus $4/$1/$12). Use the explicit surface-specific price table and recheck current terms; do not derive the long-context Fast rate by doubling the standard rate. Cursor separately publishes its own Fast price schedule and a different long-context cutoff.
What the launch benchmarks show—and what they do not
xAI’s September 21 announcement reported Grok 4.7 results across coding, engineering and knowledge-work benchmarks. The excerpt below keeps the published effort settings attached to the scores. These are xAI-reported results, not measurements from our own run.
| Benchmark | Grok 4.7 | Other scores in xAI’s launch table |
|---|---|---|
| CursorBench 4.0 | 46.3% (xHigh) | Grok 4.6 High: 40.4%; GPT-5.6 Sol Max: 41.7%; Fable 5.1 Max: 51.8%. |
| DeepSWE v1.1 | 71.0% (High effort; marked with an asterisk) | Grok 4.6 High: 65.2%; GPT-5.6 Sol Max: 72.7%; Fable 5.1 Max: 70.0%. |
The announcement’s table labels the Grok 4.7 column xHigh, but the 71.0% DeepSWE result has an asterisk and a footnote saying that score used high effort. The comparison columns use different effort labels, and the table does not include every current model. OpenAI announced GPT-6 Sol the next day, September 22; that newer model is absent from xAI’s September 21 comparison. So the table is useful as a dated, vendor-selected snapshot, not a current neutral ranking or proof that 4.7 wins across coding work. For the older Grok 4.6 query and its own benchmark context, keep the separate Grok 4.6 pricing and benchmark guide.
A fair way to evaluate it on your own tasks
Before making Grok 4.7 a default, run a small comparison on the work your team actually sends to an agent. Keep the test set and execution setup fixed:
- Choose a handful of representative tasks, such as a failing-test fix, a multi-file change and a bug investigation. Use the same repository state, task prompt, allowed tools and time limit for each model.
- Record the model surface and variant, reasoning effort, context length, tool calls and price schedule. A direct API call, a Cursor run and a Grok Build session are not interchangeable billing routes.
- Score accepted changes, test results, scope errors, human rework and elapsed time. Re-run tasks that are nondeterministic rather than treating one completion as a stable result.
- Compare cost per accepted task, including tool charges and rework, alongside latency and quality. A cheaper token rate or a vendor benchmark score alone cannot answer that decision.
This is a proposed evaluation protocol, not a test we ran for this article. The evidence checked here confirms where Grok 4.7 is listed and what its published price tiers are; it does not establish which model will perform best on your repository or workflow.
Sources and method
Product and price details were checked against the xAI launch announcement, xAI model documentation, xAI API pricing, the xAI release notes, the Cursor Grok 4.7 guide and Cursor’s current model and pricing page. The launch benchmark scores are xAI-reported. We did not run an independent model test or verify account-specific usage in Cursor or Grok Build.
