Grok API pricing for a UK business, and what Grok Bot is

Last checked: 12 Sep 2026. Every price below comes from SpaceXAI's own pages: the API pricing page at docs.x.ai (last updated 7 Sep 2026), the plan comparison at x.ai/pricing, and the Grok Bot documentation at docs.x.ai (last updated 11 Sep 2026). Prices are converted from US dollars at £1 = $1.35 on 11 Sep 2026; vendors bill in dollars and UK VAT applies on top.

Two different questions get asked as if they were one. "What does Grok cost?" usually means the API, which is metered per token and is cheap. "Can we use Grok Bot?" means a separate agent product that runs on a cloud computer and drives your applications. They are priced differently, governed differently, and carry different risks. This post separates them and gives the numbers in pounds.

Grok API pricing, converted to pounds

SpaceXAI bills per million tokens and splits each model into a short-context and a long-context rate. Once a request reaches the model's long-context threshold, which is 200,000 tokens on every current model, the long-context rate applies to all tokens in that request, not just the excess. That is the single detail most likely to surprise a finance team.

Per million tokens, converted at £1 = $1.35. Source: docs.x.ai API pricing, checked 12 Sep 2026.

Three further charges matter once an application does anything beyond plain text generation:

• Web search is billed at about £3.70 per 1,000 tool calls, on top of the tokens.
• X search is about £3.70 per 1,000 calls today, but from 21 Sep 2026 it moves to about £3.70 per 1,000 posts fetched and about £7.41 per 1,000 user profiles fetched. Every parent and quoted post counts. If real-time X data is the reason you are looking at Grok at all, model this change before you build.
• Batch processing is discounted, but the discount varies by model and several models get none. On the docs as at 12 Sep 2026, grok-4.3 and the grok-4.20 family carry 20 per cent; grok-4.6 and grok-build-0.1 are listed with no batch discount. Priority processing costs double the standard rate.

A worked example: 2,000 customer emails a month

Take a support or sales team drafting replies to 2,000 customer emails a month, at roughly 3,000 tokens per email (about 2,500 tokens of thread and context in, 500 tokens of draft out). That is 5 million input tokens and 1 million output tokens a month.

Read that table twice. The model is not the expensive part. Adding a single web search to every email costs more than the entire grok-4.3 bill for drafting them. In most real deployments the cost driver is tool calls, retries and how much context you send, not the headline per-token rate. The same finding holds across vendors, which is why we tell clients to instrument token and tool usage before arguing about model choice.

Grok subscription plans

The API is separate from the consumer and business plans. From x.ai/pricing, checked 12 Sep 2026:

SOC 2 Type I and II compliance is listed against every tier, including Free. Grok Bot has no separately published price: access comes through a SuperGrok subscription link or a paid Cursor plan, and usage resets weekly.

What Grok Bot actually is

Grok Bot is not the Grok model and it is not the API. It is a product that gives each user a persistent cloud computer with a browser, filesystem and terminal, and lets them create named Bots that work on it. From the documentation, checked 12 Sep 2026:

• Bots use connectors where a service has one, and computer use (driving the browser) for everything else. Work continues when your laptop is closed.
• Bots coordinate with each other: they run in parallel, message each other, share context in group chats and hand ownership of a task across.
• Bots learn a workflow from demonstration, save it as a skill and rerun it on a schedule.
• All of one user's Bots share the same cloud computer, including its files, browser sessions and application logins. Isolation is strict between users, not between your own Bots. Anything you put on that computer is available to every Bot you run.
• A Bot has no identity of its own. It acts as the signed-in member, so it can never hold more access than that person, and every action stays attributable to a named user.

Enterprise availability was announced on 3 Sep 2026, adding an organisation-wide enable switch, network controls with a destination allowlist, audit logs, OpenTelemetry export, SCIM and admin computer management. Those controls are Enterprise only: self-serve Teams accounts do not get them, and a team with no network policy defaults to allowing all destinations.

Our CEO's early testing (X, 11 Aug 2026) found the self-collaboration genuinely works: Bots can wake other Bots and operate as a group rather than as one assistant with a long to-do list. That is the part most agent products get wrong, and it is a real reason to look at this product rather than a marketing claim.

The questions a UK buyer should ask

The gaps here are governance, not capability. Four points, each from a vendor page rather than commentary:

1. Data residency. The Enterprise tier lists data residency as a feature, but the Grok Bot security documentation is explicit: "Grok Bot computers run in the United States today", and a written residency commitment has to be requested from your account team. If you are in a regulated sector, get that in writing before a pilot, not after.
2. The regulatory record. The ICO opened formal investigations into X Internet Unlimited Company and X.AI LLC on 3 February 2026, covering their processing of personal data in relation to Grok and its potential to produce harmful sexualised image and video content. The ICO can fine up to £17.5m or 4 per cent of worldwide turnover. The investigation concerns consumer image generation rather than the enterprise agent product, but it is a fair question at a procurement board and you should have an answer ready.
3. Hosting and sub-processors. Grok Bot runs only on Cursor-hosted cloud computers. There is no on-premises deployment, no bring-your-own-image option and no private link into your network. Egress uses shared static IP ranges common to all Grok Bot customers. Anysphere, the company behind Cursor, holds ISO/IEC 27001 and ISO/IEC 42001, and Grok Bot is in the current ISO scope.
4. CRM access. We have found no HubSpot connector for Grok or Grok Bot published by either vendor. HubSpot's marketplace currently lists first-party connectors for Claude and for Gemini. Without a connector, a Bot reaches your CRM by driving a browser under a member's own login, which means no scoped permissions, no API-level audit trail and no way to revoke the agent without revoking the person. That is a materially weaker control surface than a connector, and it is the single biggest reason we would not put Grok Bot near a customer database today.

Where it fits, and where it does not

We do not rank models and we do not have a house vendor. On the evidence above, Grok is a good fit for two things: work that needs live X data, where it has a structural advantage no competitor can match, and multi-agent work where several Bots genuinely need to coordinate. It is a poor fit where UK or EU data residency is a hard requirement, where you need scoped, revocable, audited access to a system of record, or where a self-serve Teams plan would leave you without network controls and audit logs.

The honest position is that Grok Bot is capable and under-governed. That gap is closable with process design, network policy, approval rules and a decision about what it is allowed to touch. It is not closable by buying the product and hoping.

If you are weighing this against the alternatives, our SpaceXAI and Cursor implementation page covers what we build with Grok, and the model-neutral AI implementation service covers how we choose between vendors. Grok sits alongside the other current models in our autumn 2026 decision table, and for a like-for-like comparison of the two most common shortlist entries, see Claude versus ChatGPT for business. If you want a view of your own systems first, start with a diagnostic, or Request a Quote.

Frequently asked questions

How much does the Grok API cost in pounds?

As at 12 Sep 2026, grok-4.6 costs about £1.48 per million input tokens and £4.44 per million output tokens for requests under 200,000 tokens, and about £2.96 and £8.89 above that threshold. grok-4.3 is roughly half those rates. Converted at £1 = $1.35; UK VAT applies on top.

Is Grok Bot included in a SuperGrok subscription?

Yes. SuperGrok, at about £22 a month, lists Grok Bot access alongside Grok 4.6 and connectors. Grok Bot has no separately published price, and usage resets weekly. Access also comes with paid Cursor plans.

Where is Grok Bot data stored for a UK business?

SpaceXAI's Grok Bot security documentation states that Grok Bot computers run in the United States today, and directs customers who need a written residency commitment to their account team. The Enterprise plan lists data residency as a feature, but the specifics are not published.

Can Grok Bot connect to HubSpot?

Not through a published connector, as at 12 Sep 2026. Bots use connectors where a service has one and drive a browser otherwise, so CRM work would happen through the browser under a user's own login, without scoped permissions or an API-level audit trail.

Is Grok under investigation in the UK?

The ICO opened formal investigations into X Internet Unlimited Company and X.AI LLC on 3 February 2026 regarding their processing of personal data in relation to Grok and its potential to generate harmful sexualised imagery. The investigation concerns Grok's consumer image generation, not the enterprise agent product.

What is the difference between Grok, the Grok API and Grok Bot?

Grok is the model family. The Grok API is metered per token and is how you build Grok into your own software. Grok Bot is a separate agent product: persistent named Bots on a cloud computer with a browser, filesystem and terminal, which can coordinate with each other and run work on a schedule.