Why does chatting with Grok Bot — with almost no coding work — consume so many sand-* model tokens and count them against my Other Models quota?
I am on Cursor Ultra. In this billing period my Spending page shows a large share of Other Models usage coming from sand-default and sand-automation (hundreds of millions of tokens). I use Grok Bot almost only for conversation. I rarely, if ever, ask it to write or change code, and I do not recall starting Cursor Cloud Agents from it.
Grok Bot is documented as having its own weekly included pool, separate from Cursor’s monthly Cursor Models / Other Models quotas. On the usage page, however, those sand-* rows are listed under Other Models and inflate that percentage.
Could you clarify:
Is sand-default / sand-automation Grok Bot chat and routines?
Does that usage actually consume the Other Models included quota (and block Claude/GPT), or is it only mixed into the table for display?
Why would normal chat, with little or no coding, burn this many tokens?
If this is a display issue, can Grok Bot usage be shown on its own meter so it does not look like Other Models spend?
I can share screenshots of Spending and the Grok Bot weekly usage bar if useful.
Yes - sand-default is your Grok Bot chats and sand-automation is your Grok Bot routines. These draw from Grok Bot’s own weekly pool (the Grok Bot bar on your Spending page), not your Cursor Other Models quota, so they won’t block Claude or GPT. Them showing under Other Models is just a display grouping thing on our side that we’re tracking - it doesn’t change what you’re charged.
The big number is almost all from a routine: each scheduled run re-sends the conversation’s context, and the usage page also counts cached-context reads, so totals look far larger than the actual new work. Pausing or deleting routines you don’t need (or starting a fresh conversation for a long-running one) will cut it down a lot.
One exception: if Grok Bot starts a coding agent on cursor.com/agents, that one does run on your regular Cursor plan.
What should I do now? Grok Bot has already used 45.9% + 10.2% of the quota. You said Grok Bot usage is separate from the Cursor Ultra allowance, but I currently cannot use Other Models in Cursor Ultra.
What should I do now? Grok Bot has already used 45.9% + 10.2% of the quota. You said Grok Bot usage is separate from the Cursor Ultra allowance, but I currently cannot use Other Models in Cursor Ultra.
seems that, the Grok bot spend Cursor’s other model quota, wait for offical staff explain it.
if Grok bot spend Other model quota, what does bot usage weekly stand for?
seems that, the Grok bot spend Cursor’s other model quota, wait for offical staff explain it.
I want to know, if Grok bot spend Other model quota, what does bot usage weekly stand for?
Following up. You said grok-bot chat/routines (sand-default / sand-automation) draw from the Grok Bot weekly pool, not Other Models, and that listing them under Other Models is display-only and should not block Claude or GPT.
The labels got renamed (grok-bot-default, grok-bot-automation, grok-bot-cua), but they still count against Other Models included quota. It is not display-only. Same as 7771 reported.
On my Ultra account right now:
Grok Bot weekly usage: 18% (resets Sep 3)
Other Models: 100%
Cursor Models: 68%
On-demand: disabled
This period, Other Models is mostly the bots:
grok-bot-default 362.3M (31.0%)
grok-bot-automation 558.2M (28.3%)
grok-bot-cua 78.9M (4.6%)
Because Other Models is full and on-demand is off, I cannot use Claude or GPT in Cursor. Only Cursor Models still work. The Grok Bot weekly bar still has plenty of room, so the weekly pool is not what gated me.
Please bill grok-bot-* only against the Grok Bot weekly meter, not Other Models included quota. Happy to share more screenshots if useful.
Hey @Daniel_kochav and @7771 you’re right and my earlier reply was incomplete, sorry. It comes down to the type of Grok Bot usage:
Conversations draw from the Grok Bot weekly pool and don’t touch Other Models (that part was right).
Routines/automations and browser actions (anything Grok Bot runs as a cloud agent) run on your Cursor plan, so they count against Other Models and can block Claude/GPT even while your Grok Bot weekly pool has room. That part isn’t display-only; I had it wrong.
It’s an issue we’re tracking; no timeline yet. To unblock now: pause or delete the heavy routines/automations (grok-bot-automation / grok-bot-cua are the main drain), or enable on-demand in Settings (paid) if you need Claude/GPT before your quota resets. For an account-specific review, email [email protected].
