Grok Bot limits: every reported ceiling and undisclosed allowance

How many Bots can you create?

Fifty Bots and group chats per account, reported. That figure, and every other countable ceiling on this page, comes from a Daily Dose of Data Science teardown published in launch week. As of 22 August 2026 it is the only place any of these numbers appear in writing. xAI has published none of them, has confirmed none of them, and can change any of them without an announcement.

Note that "Bots and group chats" is a combined pool rather than two separate budgets, per the same source. If that is right, an account running a dozen group chats has a dozen fewer Bots available, which changes how you would organise a team of them.

What is the weekly token allowance?

Unpublished, for every tier. This is the single most important unanswered question about the product and it has no answer at all.

eesel’s pricing teardown documents the shape without the number: there is a weekly token allowance, its size is not publicly disclosed, and past it you are billed at raw model rates. Nothing on x.ai, nothing in the Grok Bot documentation, and nothing on Cursor’s help pages puts a figure on it. Cursor’s help page on SuperGrok gets closest and stays qualitative: "SuperGrok Heavy grants the highest linked Grok Bot usage" and "SuperGrok Plus grants generous linked Grok Bot usage, below the Heavy tier".

Read that carefully. It ranks two tiers against each other without telling you what either one contains. You can learn that Heavy is more than Plus. You cannot learn whether Plus covers your workload, which is the only question a buyer has.

Is there a spend cap on Grok Bot?

No. eesel’s review states it plainly: "There is no Grok Bot-specific spend cap yet." Overage past the undisclosed weekly allowance is billed at raw model rates.

Three things compound to make this worse than a normal metered product.

One beta tester quoted in eesel’s teardown summarised the experience: "I’ve used more tokens this month than not this month. That’s not a typo." That is one person’s anecdote rather than data, but it is the only first-hand account of consumption anyone has published.

Our full breakdown of the four cost layers, and what each access path costs, is in the Grok Bot pricing guide.

Can you trust the usage meter?

Not at launch, and nobody has published a re-test. eesel’s review, dated 12 August 2026, documents a metering display showing 48% consumed as 0%. That is not a rounding error or a lagging refresh; it is the gauge disagreeing with the account by half its capacity.

A wrong meter is a specific class of problem on an uncapped product. On a capped plan, a broken gauge is annoying: you hit the ceiling and stop. On an uncapped one with an undisclosed allowance, the gauge is the only instrument you have, and if it under-reports you cross into overage without any signal that you have.

What does "always on" actually mean here?

It means the machine is xAI’s and it stays up. It does not mean each Bot has its own machine, and the difference is documented.

From docs.x.ai/grok-bot/overview: "All of your Bots use the same persistent cloud computer. They share files, browser sessions, and app logins, which makes handoffs possible without repeating setup." And: "Each Bot gets its own screen on that computer, so several Bots can use browser and desktop tools in parallel without getting separate security boundaries."

So parallelism is real: several Bots can drive several browsers at the same time, which is more than most agent products offer. What is not real is isolation. Fifty Bots on one account is fifty personas on one machine with one set of credentials, and the documentation asks you to "treat a login or file placed on the computer as available to all of your Bots".

Nothing published describes what happens when parallel Bots contend: two routines writing the same file, or two Bots driving the same authenticated session in different tabs. There is no documented lock, queue, or arbitration. The security page works through the consequences of the shared boundary; the routines page covers the scheduling side.

What limits does the plan you are on impose?

Access is the first limit, and it is the one that is actually documented. As of 22 August 2026 the get-started documentation lists eligible plans as "SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium", and a Cursor account is required on every path.

Two rows deserve emphasis because they catch people out. SuperGrok Team and SuperGrok Enterprise do not grant Grok Bot usage. Cursor’s help page says so directly: "Only individual SuperGrok Plus and individual SuperGrok Heavy grant Grok Bot usage", and "SuperGrok Team and SuperGrok Enterprise plans are not supported". The naming is a trap, because "SuperGrok Team" sounds like the team plan that would obviously include it and "Cursor Teams" is the one that actually does. The enterprise guide covers the b2b path properly.

Which limits should change your plan?

Ranked by how likely each is to matter in the first month.

• The uncapped meter, first. Assume there is no ceiling on your bill and behave accordingly: no routine that can loop, an explicit exit condition on every one, and a billing alert on the card if your provider supports it.
• The reported 20-run history, second. This limit was reported by Daily Dose of Data Science and remains unconfirmed by xAI. Make each routine write its own log to somewhere you control. The product will not keep your history for you.
• The shared computer, third. One credential set, no boundary. Decide what goes on that machine as if every Bot you will ever create can use it, because per the docs it can.
• The reported 10-minute recording limit, fourth. It was reported by Daily Dose of Data Science and remains unconfirmed by xAI. It only bites when you are teaching. Split long workflows into skills at every point where the procedure waits.
• The reported 50-Bot ceiling, last. It was reported by Daily Dose of Data Science and remains unconfirmed by xAI, and almost nobody will reach it. Group chats reportedly consume the same pool.

If the uncapped meter is the thing that stops you, the structural fix is to run agents somewhere the cost is a subscription you already hold rather than a meter you cannot see. Continuum runs Claude Code, Codex, Cursor, Grok, Gemini, and OpenCode on hardware you own under your existing subscriptions, shows live five-hour and weekly quota gauges next to the work, and records spend by repository so a runaway session is visible while it is running rather than at the end of the month. The app is free; hosted inference is optional and starts at $25 a month with a published weekly allowance, which is the part Grok Bot does not give you at any price.

Questions people ask

Keep reading

Continue through Grok Bot with Is there a Grok Bot free trial? What xAI announced, and what it did not.