I’m a cursor pro plan user.
I tried use Grok Bot, it seems pretty good.
but when I see the dashboard, I’m supprised, it spend my model quota. I can’t select which model for bot use.
I thought Cursor Bot would have a separate usage pool,
but only after using it did I realize that it shares the same usage pool with Cursor.
I believe this should be clearly communicated to users.
Hey, thanks for the feedback. About usage, this is expected behavior. Grok Bot is billed to your Cursor account and uses your plan’s weekly usage on Pro, not a separate limit. After your weekly included usage runs out, if on-demand is enabled, usage can keep going from your shared on-demand spend. Details are here: Plans and billing | Cursor Docs
About choosing a model for the bot, there isn’t an option for that right now. You can’t manually pick the model.
A couple of related threads, if you’re interested:
And here’s a feature request for a separate allowance. You can vote on it here: Allow using Other Models allowance for Cursor Models (Grok / Composer)
this is the email i’ve got, it clearly says grok bot uses a SEPARATE usage pool:
Grok Bot is now included with your Cursor Pro plan. Your plan now comes with a dedicated Grok Bot usage pool, separate from your Cursor usage.
Grok Bot lets you give real work to AI teammates. Download Grok Bot for desktop or iOS and hand Bots work you already do
This was a bit confusing for me, but asking GrokBot questions about it cleared it up for me. GrokBot is a persistent agent. The amount that you can interact with GrokBot is limited (weekly rolling limit). When GrokBot performs actions on your behalf (e.g. “Can you add this feature to the whatever-app repo and submit a PR?”) it then calls other agents on your behalf (you can tell it to use a specific one or just let it pick, I tell it to use Grok 4.6 non-fast). That usage then counts against your Cursor usage the same as if you had made the request yourself.
Thanks for the message, @m4.5. I need to correct my previous reply, sorry for the confusion. My note about the weekly allowance was correct. On Pro, Grok Bot has its own separate included weekly pool, separate from your Cursor plan usage. My wording in the last message was wrong.
Important clarification: separate only applies to the included weekly limit. Once that weekly pool is used up, any further Grok Bot usage, if you have on-demand enabled, will start billing from the shared On-Demand pool, the same pool used for the rest of Cursor usage. So it is not fully isolated after the weekly limit is exhausted.
To avoid unexpected charges: in the dashboard on cursor.com go to Dashboard > Usage / Billing and set a dollar limit for On-Demand usage-based. If you want to fully prevent any paid usage, set the limit to $0, then there will be no spillover to On-Demand.
@Warren_James described the general setup well. Grok Bot is a persistent agent with a weekly limit. The only add-on is that included Grok Bot usage comes from that separate weekly pool, and only the overflow moves into Cursor usage after it runs out, as noted above.
One more note about the dashboard: it currently does not visually split Grok Bot usage vs Cursor usage, and it does not break down On-Demand by product.
There is no timeline yet for a warning tooltip or a per-product breakdown. If I get an update, I’ll post it here.
When Grok Bot is running, it uses AI models to perform its tasks—for example, it uses Sonnet in this case, which is counted against my “Other Model” usage quota. However, Grok Bot also has its own usage quota, which makes me a bit confused.
In other words, Grok Bot consumes from two usage quotas at the same time:
- Other Model usage quota
- Grok Bot’s own usage quota
update:
please ignore the post above the split line. There is a answer I find:
thank for @deanrie 's reply. 
For me it draws from the included usage, both Cursor and Other models. While I was running only Grok Bot, with disabled On Demand, and without using cloud agents, and without using Cursor IDE (I even disabled all my sessions to make sure my session didn’t get hijacked or something) - Grok Bot while performing automation on its Desktop was using:
- claude-opus-5-thinking-low
and - cursor-grok-4.6-high-fast
It used 7.1% of my included allowance of Cursor Models during a short run and also around 50M of Claude Opus Low which wasn’t counted at first, but after a few days it started to recalculate and the same amount that was showing 0.7% usage now shows 12% usage and it’s all Grok Bot run.
My Grok Bot weekly was (and still is) below 40%.
I have opened a support case for that (T-F21229) which has proofs.
Avoid Grok Bot if you don’t watch your usage. @deanrie can you take a look at this?
Now, every time I use Composer 2.5 in my IDE the Claude Opus Low percent usage is also growing in my Other Models pool - but not the token usage. It looks like it’s recalculating (and adds on top of Claude) every time I use my IDE. I am already past my included allowance and I am using $20 included API usage.
This is super weird. Nothing is growing when I don’t use anything. Claude Opus Low and Grok High Fast entries stopped immediately showing up when I stopped Grok Bot.
I’m seeing the same issue on my Pro+ account.
I opened support ticket T-F33183. Cursor Support explicitly confirmed that my grok-bot-default and grok-bot-cua rows are included in the monthly Cursor Models and Other Models percentages shown on the Spending page.
In my case:
grok-bot-cua: 15.8%github_bugbot: 6.1%gemini-2.5-flash: 0.2%
These add up to approximately 22.1%, matching the 22% Other Models used shown on Spending.
Likewise, grok-bot-default contributes 8.0% to the entries that add up to my 24% Cursor Models used.
At the same time, my separate Grok Bot weekly allowance shows 82% used, and On-Demand is disabled.
Support also confirmed that Grok Bot’s weekly included allowance is supposed to be separate from the monthly Cursor Models / Other Models pools.
However, support said they cannot determine whether these monthly percentages are only a display/accounting issue or whether they are also used for actual quota enforcement, and they could not escalate the case to engineering.
They specifically told me that if my monthly Cursor Models or Other Models later hit a hard stop while I was only using Grok Bot, I should reply with that outcome so they could investigate further.
That effectively leaves users to discover the answer by exhausting their paid monthly allowance, which does not seem like a reasonable way to verify whether quota enforcement is working correctly.
This now appears to affect at least multiple independent accounts, given the similar reproduction reported here under T-F21229.
@deanrie could you please help escalate T-F21229 and T-F33183 to the usage/billing engineering team and clarify whether Grok Bot usage is actually counted toward monthly quota enforcement?
Just an update from my support ticket: I finally got a detailed confirmation from Cursor that the Grok Bot usage showing up under Cursor Models / Other Models is a display issue, not actual consumption of the monthly Cursor allowance.
Support checked the internal tracking and confirmed that Grok Bot requests are recorded against the separate weekly Grok Bot allowance. They are not passed to the counter responsible for the monthly Cursor allowance.
They also explained what seems to be causing the confusion: the Included Usage panel aggregates usage events by model, but currently does not properly filter them by the product label that distinguishes Grok Bot usage from normal Cursor usage. Because of that, Grok Bot usage can appear inside the Cursor Models and Other Models breakdowns even though it was not actually deducted from those allowances.
They verified this against my own account as well. My Cursor Models allowance was genuinely exhausted on Aug 29, almost entirely from Composer 2.5 usage. According to their calculation, if my Grok Bot usage had actually been deducted from the same allowance, I would have hit the limit on Aug 27 instead.
There are still two things support is investigating for me:
- what the real Cursor vs Grok Bot split should look like in the affected usage panel;
- and, importantly, exactly what happens when the displayed Cursor usage reaches or goes past 100% while part of that displayed usage is actually misattributed Grok Bot usage.
So at this point I have independently verified that Grok Bot is not actually eating the monthly Cursor allowance, but I am still waiting for confirmation about the behaviour around the 100% threshold (I might actually check it myself soon as I am approaching the limit 
).
I’ll update the thread again when I get a definitive answer on that.