Grok 4.7 Is Out: New Features, Pricing and How to Try It

Grok 4.7 launched on September 21, 2026, with access through Cursor, Grok Build and the xAI API. Its standard API pricing starts at $2 per million input tokens and $6 per million output tokens, the same base rates as Grok 4.6. The new model targets coding, agents and professional work.

The most useful starting point is the product you already use. Cursor users can check the model picker; Grok Build users get 4.7 as the documented default; developers can request grok-4.7 through the API. A free Grok Build entry point is available, but it does not include Grok 4.7 Fast.

Updated September 21, 2026. This guide is based on xAI’s launch announcement, model documentation and published pricing. We have not performed a new account-by-account rollout test or a paid Grok 4.7 hands-on review. Featured image: xAI’s official Grok 4.7 launch artwork, from the announcement.

For the evaluation tables and comparisons with GPT, Claude and Gemini, read our Grok 4.7 benchmarks and specifications analysis.

What is new in Grok 4.7?

In its release announcement, xAI describes a larger base model, longer reinforcement learning and more training on difficult work that takes hours. It reports better self-checking, context management, documents and presentations, plus a redesigned safeguard system.

For a user, the question is whether the model can stay with a job until the output is useful. A coding agent should find the relevant files, implement a change, run the tests and investigate failures. A research assistant should keep citations attached to the right claims. A document task should end with an editable file that meets the brief.

Those are sensible things to test first. The launch claims describe intended improvements; they do not establish that every request will finish correctly.

The official model overview confirms a 500,000-token context window, a May 2026 knowledge cutoff, text and image inputs, and text output. It supports low, medium, high and xhigh reasoning, with high as the default. The capacity is useful for large inputs, but it does not make context management unnecessary.

Where to use Grok 4.7 today

xAI’s availability documentation identifies the following routes. Availability in a product does not mean unlimited included usage.

For consumer Grok chat, do not assume every app surface or subscription has switched to 4.7. At our check, the consumer plan page still named Grok 4.6. Its broad plan description is not confirmation that every web, mobile or X chat session is now running the new model. Check the model shown in your account.

Grok 4.7 pricing: standard, cached and long context

The official API rate card distinguishes standard requests from long-context requests. All figures below are US dollars per million tokens.

The higher tier applies to all tokens in a qualifying request, not just the part above the threshold. An agent’s repeated requests, billed reasoning and tool usage also contribute to cost.

What does a request cost in practice?

These examples use standard global API rates. Output means total billed output, including reasoning. They are calculations, not measured workloads; separate tool charges and provider fees are excluded.

For example, 50,000 × $2 ÷ 1,000,000 + 5,000 × $6 ÷ 1,000,000 = $0.13. A ten-request agent task with that same usage on every request would cost $1.30 before tools. Count every request when comparing the cost of a finished job.

Two billing details matter for integrations: the US regional endpoint adds 10% to token prices. And xAI’s tool pricing notice schedules X Search’s switch to $5 per 1,000 posts fetched and $10 per 1,000 profiles fetched for September 21, 2026 at noon Pacific. Search results can therefore cost more than a single search call.

What is Grok 4.7 Fast?

The September 21 API release notes describe Fast as the same model served through Cursor and Grok Build. It is not available on the public xAI API. The overview also excludes it from Grok Build’s free tier.

xAI advertises twice the output speed. Faster output can make interactive work feel better, but it does not halve the time spent waiting for a web search, build or test suite.

For short-context Fast requests, the listed rates are $4 input, $1 cached input and $12 output per million tokens. The pricing page’s Fast long-context table lists $6, $1.50 and $18 respectively, despite describing Fast more broadly as twice the standard token rates. Those long-context numbers do not equal twice the standard long-context row. Use the explicit product rate shown before starting a large job; do not extrapolate the short-context multiplier.

Public API Priority Processing is documented separately. It should not be confused with a public API model called Grok 4.7 Fast.

How to try it in Cursor or Grok Build

1. Open your existing environment. In Cursor, look for Grok 4.7 in the model picker. For Grok Build, use the installation or sign-in path on xAI’s official Build page.
2. Confirm the variant and usage terms. Standard and Fast have different charges. A model appearing in a picker does not tell you how much usage your subscription includes.
3. Give it one bounded task. Choose a bug, document or small feature with an outcome you can check. Supply the relevant files and a clear definition of completion.
4. Review the result. Check the diff, run the tests, open the generated file and inspect citations where relevant. Record the time and usage before comparing it with your usual model.

A useful first coding request is: “Find the cause of this failing test, make the smallest appropriate fix, run the relevant tests and explain what changed.” Start in a branch or disposable copy so you can inspect the edits easily.

For a document, provide a source packet and ask for a short recommendation with each factual claim linked to its source. Judge whether it used the supplied evidence correctly and whether the resulting document needs substantial repair.

How developers can access the API

The model identifier is grok-4.7. Create a key in the xAI console and check your account’s billing and limits. Send the JSON body below to POST https://api.x.ai/v1/responses using the authentication setup in xAI’s official API examples. Submitting a request incurs API charges.

{
  "model": "grok-4.7",
  "reasoning": {"effort": "low"},
  "input": "Explain why sorting numbers in JavaScript needs a numeric comparator."
}

We checked this example against the documentation; we have not run it as a paid test. The JSON response contains the model’s answer in its output items. This request asks for an explanation; it does not connect the model to your files, terminal or search tools.

Three integration details to get right

1. Set reasoning deliberately. The example uses low effort for a small task. The default is high; xhigh spends more time thinking. Reasoning cannot be disabled, so a short visible answer can still include billed reasoning tokens.
2. Make cache reuse possible. xAI recommends a stable prompt_cache_key for each Responses conversation, or x-grok-conv-id for Chat Completions. Keep reusable prompt prefixes consistent and inspect reported cached tokens. A cache key helps route requests; it does not guarantee every input token gets the cached price. See the caching guide.
3. Preserve reasoning between turns. Grok 4.7 Responses output includes reasoning.encrypted_content automatically. If you manage conversation history yourself, pass the returned reasoning items back unchanged in the next request’s input. The reasoning documentation explains the behavior; Chat Completions is unaffected.

Also check the model detail page before planning asynchronous bulk jobs: it lists Batch API as unsupported. A general Batch API feature on the platform does not guarantee that every model accepts it.

Is Grok 4.7 worth switching to?

For existing Grok 4.6 coding users, a trial is easy to justify. The standard rate card is unchanged, and xAI reports improvements in coding and professional work. Teams using Claude or GPT should compare a representative set of tasks before changing their default.

For a useful comparison, give each model the same bug fix, document task and longer research task, with the same files and completion criteria. Record whether the result passes, how much correction it needs, elapsed time and total cost. Keep the reasoning setting and available tools beside each result. Three tasks are a screening exercise, not a benchmark, but they can expose an expensive mismatch before you change a team’s default.

Use the benchmark comparison to choose which alternatives to test. Use your own completed work to decide which one to keep. If subscription limits matter more than token rates, our AI coding plan limits guide explains why those are different purchasing questions.

Frequently asked questions

Is Grok 4.7 free?

xAI advertises a free starting point through Grok Build. Limits still apply, Fast is excluded from the Build free tier, and public API usage is separately priced. This does not establish unlimited free 4.7 access across every Grok app.

Can Grok 4.7 create images or videos?

The model accepts images but outputs text. Image and video generation are separate xAI products. A Grok application can combine those capabilities without making them outputs of the Grok 4.7 text model itself.

Is Grok 4.7 Fast a smaller model?

xAI describes it as the same model on faster serving infrastructure. Its availability and billing differ from the standard public API offering.

What should I do if I cannot see Grok 4.7?

Check the product’s model list and your account’s access details. The launch confirms particular routes; it does not guarantee that every old chat, subscription screen or third-party integration has updated. Use the documented API or Build route if it is available to your account.