Grok Bot weekly usage limit: why it runs out and how to stretch it

Your Grok Bot allowance resets weekly, is measured in agent steps and tokens rather than messages, and works differently depending on whether you are on The X Subscription or an older Cursor plan. Most "it ran out in two days" stories trace back to long conversations, Bots waking other Bots, and routines that fire too often. Here is how the meter works, what burns it, and which caps actually stop spending.

At a glance#

The X Subscription launched on October 12, 2026. Existing Cursor and SuperGrok plans continue for current subscribers, so check which one you are on before applying any advice below. Our pricing guide covers the plans themselves.

Which plan you are on changes everything#

On The X Subscription, SpaceXSI's docs describe one weekly pool per paid seat, shared across eligible Grok and Cursor products, with usage rates that vary by product. A Cursor staff member confirmed on October 20 that there is no separate Grok Bot allowance on Premium, Super or Ultra. On a team, each seat has its own pool and the team shares one prepaid balance.

On the legacy plans, Grok Bot had its own weekly bucket. Cursor staff confirmed in September that Grok Bot chats and routines do not touch your Cursor plan quota, with one exception: a Cursor Cloud Agent that a Bot launches bills to your Cursor plan, and on a team, to your team seat.

The catch on legacy plans is overflow. Staff first told one Pro user in August that Grok Bot used Cursor quota, then corrected themselves: the weekly pool is separate, but only up to its limit. After that, on-demand usage comes from the same shared pool as the rest of your Cursor spend. If Grok Bot seems to eat into Cursor spend early, check which pool the usage screen charged.

Allowances also do not stack. Cursor's billing page says a Cursor plan plus a linked SuperGrok or X Premium+ subscription gives you whichever grants more, not both. After upgrading a linked SuperGrok tier, allow up to 24 hours for the new allowance to appear.

A display bug that looked like double billing#

In September, grok-bot-default and grok-bot-cua rows appeared in the Cursor Models breakdown on a Teams Spending page. Cursor staff called it a display bug and said the rows were removed by September 11. The headline percentages had stayed Cursor-only throughout.

Why your tokens disappear so fast#

Every turn re-reads the Bot's entire conversation, and the product only auto-summarizes near the model's limit. In one account staff reviewed, the busiest Bots were re-reading roughly 80,000 to 90,000 tokens per step, and about a third of all turns were Bots waking other Bots.

That is why a single one-line reply can show 50,000 to 165,000 tokens. Staff explained on October 14 that the Tokens figure counts the Bot's instructions, memory, tool list and recent chat, and that most of it is cached context billed at a lower rate. Hover over the number on the usage page to split it into input, output, cache read and cache write.

The usual culprits#

• Parallel fleets. One user lost 42 percent of a week on day one across six Bots.
• Chatty routines. Cursor's routines help warns that an hourly schedule, a short interval or a Slack trigger on a busy channel can use a week of usage in a day.
• Long browser sessions. Twenty minutes of scrolling costs far more than twenty short replies.
• Retry loops. A Bot hammering a stuck page or a rate-limited connector.
• Debugging during outages. On October 8, one user had over 90 percent used with five days left after asking Bots to investigate failed routines. Check the status guide first.

Run caps will not save you: since 0.75.5 your routines can run up to 300 times an hour in total and each routine up to 200 times an hour. One user reported spending $100 of on-demand usage in under a day, and another agreed on October 27.

How to make a week last#

There is no model picker, so you cannot route small jobs to a cheaper model. Your levers are structural.

1. Start a fresh chat per task. A thread that grows forever costs more on every reply. Our shared computer guide explains how to open a new chat with the same Bot.
2. Move standing instructions into the Bot's profile so a fresh chat loses little.
3. Tell Bots to message each other only when action is needed. Bot-to-Bot traffic was a third of turns in one audited account.
4. Space out routines and give each one a stop-if-nothing-changed instruction. See routines not running for writing routines that fail loudly instead.
5. Trim each Bot's background. One user cut per-call size to about a third with a leaner setup and fewer event-triggered wake-ups.
6. Shrink Cloud Agent jobs with an AGENTS.md map of each repo and a .cursor/CLOUD.md file, as staff suggested.
7. Update the app. The October 11 release (0.69.0 to 0.70.2) makes Bots wait for a rate-limited connector's retry delay instead of looping.

Copy-paste routine guard#

Before doing any work, check whether [source] has changed since your last run.

If nothing changed, reply with one line: "No change since [last run time]" and stop.

Do not message other Bots unless a result needs their action.

Keep this run under [N] steps; if you hit that, stop and report where you got to.

Set the caps before you need them#

Do this before the first routine runs unattended. On a legacy self-serve Cursor Teams plan, on-demand is enabled by default.

On The X Subscription#

SpaceXSI's docs list four controls: the weekly pool, a prepaid balance, a monthly additional usage limit, and Auto Top-up, which stays off unless you turn it on. Set them under Additional usage credits at grok.com/dashboard. Adding credit does not raise the monthly limit, and spend alerts warn without stopping anything.

On legacy Cursor plans#

1. Open Settings → Usage & Billing and find On-demand monthly limit (choose Enable if it is off). On the web it is Spending, Monthly Limit at cursor.com/dashboard. If you subscribed in the iOS or Android app, set it on the web.
2. Either disable on-demand, so Bots stop when the allowance ends, or set a limit you can afford. A run already in progress can finish past it.
3. Note today's level and the reset time, shown since 0.63.0. Since the October 11 release the percentage is labelled as used: 80% means 80% gone.
4. Run one Bot on one bounded task and re-read usage to learn what a single run costs.
5. Add routines last and check usage after their first unattended week.

Reading the warnings#

• Usage limit notice instead of "Bot failed to respond": shown since 0.63.0 when you run out.
• Upgrade, Set Spend Limit or Request Upgrade: the limit banner's options since 0.59.0.
• "Ask your Grok Business admin to connect your team": since 0.75.5 (October 23), your Grok Business seat does not cover Grok Bot yet.
• Banner that will not go away: it returns after dismissal; since 0.78.1 (October 27) it closes on its own once you are back under the limit.
• Blank Usage & Billing: since 0.77.0 (October 26) it keeps the last reading with Try again.
• "Deleted bots, group chats and other people's bots": the catch-all row in a Bot's usage breakdown since 0.78.1.

On a shared Bot, the docs say a reply uses the usage of the person it is answering. "Your usage limit has been reached" means your own allowance ran out; "This Bot's owner has reached their usage limit" means that reply was billed to the owner.

Common mistakes#

• Assuming a monthly limit on a legacy plan halts a Bot mid-task. It does not.
• Running one ever-growing chat per Bot for weeks.
• Letting a change-watch routine wake a Bot on every pull request or Slack message.
• Asking Bots to debug themselves during a declared incident.
• Buying a second subscription in the mobile app when your Cursor account already includes Grok Bot.

Our take#

Treat the weekly pool as a budget for steps, not chats. Set your hard stop first, keep conversations short, and make every routine prove there is new data before it does work. If you are still burning a week in days after that, the problem is fleet design, and our troubleshooting matrix is the next stop.

Frequently asked questions#

Does Grok Bot have its own usage allowance?#

It depends on your plan. On The X Subscription (Premium, Super, Ultra) there is no separate Grok Bot allowance: it draws on the same weekly pool as Grok and Cursor, per a Cursor staff reply on October 20. On the older Cursor plans, Grok Bot has its own weekly pool, but once that runs out, extra use bills from the shared Cursor on-demand pool.

Is the Grok Bot spending limit a hard stop?#

On The X Subscription, the monthly additional usage limit stops paid extra usage, and extra usage also stops when your prepaid balance is empty; spend alerts only warn. On legacy Cursor plans, the On-demand monthly limit is not a hard stop mid-run, so a working Bot can finish past it. Cursor staff suggested a $0 on-demand limit if you want a hard stop.

How big is the weekly Grok Bot allowance?#

No absolute number is published. On The X Subscription, x.ai and Cursor pricing give Premium 4x the usage of Plus, Super 15x and Ultra 50x, while the x.ai/bot panel puts Super at 3.5x and Ultra at 12x Premium. Cursor staff said on September 22 there is no tier above Ultra and no published dollar figure.

When does the Grok Bot allowance reset?#

Weekly, but Cursor's billing page does not name a fixed day. Since version 0.63.0 (September 29), Settings, Usage and Billing shows your included usage and when it resets. On The X Subscription, the live usage view at grok.com/dashboard shows the next reset date. Update the app if you see no reset time.

Why does one short reply show 50,000 tokens or more?#

Each row on the usage page counts everything the model read on that call: instructions, memory, tool list and recent chat. Cursor staff said on October 14 that most of it is cached context billed at a much lower rate. Hover over the token count to split it into input, output, cache read and cache write.