← All posts

Guides

xAI Pricing: What Should Enterprises Budget for Grok?

Budget Grok API tokens, Web Search calls and X Search items separately. See a worked example and what to confirm in a Token360 enterprise quote.

For a Grok API research workflow, budget for model tokens and the specific search tools used. Web Search calls and X Search content have different billing units, so one user question is not one predictable search charge. Keep this API estimate separate from a Grok app subscription and from any negotiated Token360 enterprise quote.

This guide addresses a specific enterprise purchase: using Grok through an API for research or analysis that may retrieve web pages and X content. It explains the public billing units and a transparent planning example. It does not evaluate model quality, promise a fixed cost per answer, or confirm that every tool is available through Token360.

Are Grok subscriptions and API usage billed together?

No. The official account FAQ distinguishes Grok billing from API billing even when the same account is used to sign in. An enterprise building an application should therefore establish an API usage budget, rather than treating a consumer app subscription price as the price of its production model calls.

The API account FAQ directs API billing to the console and Grok billing to the app or the relevant app store. Source checked November 20, 2026. This separation answers the purchasing question; it does not require comparing every consumer subscription plan.

Before requesting a quote, identify what employees or software need to do. People using a chat application and a service automatically generating thousands of research summaries are different purchases. A shared product name does not make their entitlements or billing units interchangeable.

For clarity, this article uses the search term “xAI pricing” for the API at api.x.ai. The current official documentation carries SpaceXSI branding. The rates below refer to the linked API documentation, not a Grok app plan or an X subscription.

What are the public Grok token rates?

The model and context range determine the applicable text-token rates. For Grok 4.7, the public rate card separates ordinary input, cached input and output, with a higher-price band for long-context requests. The table below is a dated public API reference; it is not the price of a Token360 enterprise offer.

Grok 4.7 public text API rates · USD per million tokens · November 20, 2026
Prompt sizeOrdinary inputCached inputOutput
Below 200,000 tokens$2.00$0.50$6.00
At least 200,000 tokens$4.00$1.00$12.00

The official API pricing page says the long-context rate applies to all tokens in a request once its prompt reaches the threshold. The Grok 4.7 model page also lists the input, cached-input and output rates. Sources checked November 20, 2026. This example uses the standard global rates, not a regional or accelerated-service price.

For enterprise planning, preserve the distribution of request sizes instead of averaging everything into one large monthly token count. A workflow with occasional large research contexts needs those requests assigned to the correct band. A small average does not prove that every request qualifies for the lower row.

Do not assume the cached rate applies to repeated business questions. Use the actual billable cache category reported for the request. If caching has not been measured, keep the expected benefit separate from the approved baseline rather than presenting it as an established saving.

Are Web Search and X Search charged per API request?

They are not both priced per user request. The public schedule prices Web Search by tool calls, while X Search charges for fetched posts and user profiles, in addition to token costs. A research task can therefore have a small number of submitted questions but a larger amount of billable retrieval work.

Selected search meters · USD · November 20, 2026
Search activityPublic rateQuantity to record
Web Search$5 per 1,000 callsBillable Web Search tool calls
X Search: posts$5 per 1,000 postsBillable posts fetched
X Search: profiles$10 per 1,000 profilesBillable user profiles fetched

Sources: API tool pricing and the X Search documentation, checked November 20, 2026. The pricing documentation counts posts returned by searches or thread fetches, including parent and quoted posts, toward the post meter. Do not substitute the number of final citations for the billed retrieval count.

Suppose a research question asks the system to compare discussion around two product launches. The final answer might contain only a short summary, while the retrieval work examines multiple posts or profiles. That is why “we expect 1,000 questions” is not yet a complete tool-cost estimate.

For a purchasing brief, distinguish the required research outcome from its estimated retrieval volume. Engineering should estimate or measure the billable quantities; the business owner should decide whether the resulting research is useful. Neither the number of fetched posts nor the amount spent establishes the quality of the conclusion.

What would an illustrative Grok research workload cost?

Under the explicit assumptions below, the selected public rates produce a $70 subtotal for a hypothetical monthly workload. The total combines text tokens, Web Search calls and fetched X items. It is arithmetic for planning, not a measured customer bill, a per-question guarantee or a Token360 enterprise quote.

Assume all model requests stay below the long-context threshold. Aggregate billable usage is 10 million ordinary input tokens and 2 million output tokens, with no cached input. Search usage is 2,000 Web Search calls, 5,000 X posts and 300 X profiles. Assume the token totals already include all billable text-token usage from the workflow, rather than only the final answers.

Hypothetical monthly subtotal at the dated public rates
CategoryCalculationAmount
Ordinary input10 million × $2 per million$20
Output2 million × $6 per million$12
Web Search2,000 ÷ 1,000 × $5$10
X posts5,000 ÷ 1,000 × $5$25
X profiles300 ÷ 1,000 × $10$3
Subtotal$20 + $12 + $10 + $25 + $3$70

This excludes other tools, media processing, regional or accelerated-service adjustments, taxes, and any separate contractual charges. It also excludes any unrecorded usage. Replace the quantities and applicable rates before treating the example as an operational budget.

One useful sensitivity check isolates the retrieval meter: doubling the fetched posts from 5,000 to 10,000 adds $25 in post charges at the same rate. It does not establish that the full workflow increases by only $25; additional retrieval may also change token consumption. Keeping that distinction visible prevents a tool-only calculation from being mistaken for an all-in forecast.

What should you verify before approving the budget?

Verify the model, request-size band, token categories and each search meter using representative work. Keep the final response and its usefulness separate from the usage record. The aim is to establish what your application consumes and whether the result is acceptable, not to infer a stable per-answer cost from one successful demonstration.

A small validation record should retain the business task, model and endpoint, ordinary and cached input where reported, billable output, search quantities, and the applicable price date. Record unsuccessful attempts and retries separately where they generate billable usage; do not assume every error or every retry is charged identically.

Reconcile the chosen measurements with the account’s billing record. The official API billing guide describes prepaid credits and points to its usage explorer. Source checked November 20, 2026. Buying credits funds usage; the credit purchase alone does not establish the cost of a particular research task.

For Token360 usage, the API billing reconciliation guide owns the detailed reconciliation process. This article’s contribution is the Grok-specific distinction between token consumption, tool calls and fetched items. A representative test has not been run for this article, so no observed accuracy, latency or cost-per-accepted-answer result is claimed.

What should Token360 quote for a Grok deployment?

Ask Token360 to confirm the exact Grok model, endpoint and search-tool requirements before requesting rates. Token360’s enterprise quotes are tiered: higher usage receives lower quoted rates, with no separate service fee. The model route, eligible usage and any tool charges still need confirmation in the actual offer.

A request for “Grok access” is too broad for this workflow. Specify whether the application only needs text generation or also needs native Web Search, X Search, file retrieval or media analysis. Model access does not establish support for every upstream tool. This article does not confirm Grok 4.7 or native search-tool availability through Token360.

Send the expected monthly token quantities, request-size distribution, search volumes and peak demand. Ask which billing categories the enterprise tier changes and which, if any, remain separately priced. Do not apply a token discount to every search item unless the quote explicitly covers it. The volume-discount guide explains those commercial distinctions.

Token360’s enterprise offering covers model procurement across text, image, audio and video with custom pricing scoped to production requirements. Its billing documentation distinguishes the active route price from an upstream public reference. No separate service fee means no separate service fee; it does not mean free inference or a guaranteed saving against direct access.

For a comparable purchase, request a written model-and-tool scope and a worked estimate using your measured quantities. Use the complete LLM quote-comparison framework if assessing more than one supplier. The decision should rest on an offer that covers the research workflow you need, not only an attractive headline token price.

Written by Yuki Yu for Token360. Public documentation checked November 20, 2026. Quantities and totals are illustrative; no live test, negotiated discount or model-quality comparison is claimed. Recheck the current rates and availability before purchase.

  • Enterprise AI
  • API

Bring your next idea to life.

Create a video scene in Creator, or explore models for images, audio, and text to find the right capabilities for your project.

Get started