Hi someone please shed some light on this? How is it possible after just few days since my last reset that I have expended so much tokens by 80% already when using the bugbot tool - enabled for daily run to perform vulnerability scan?
In total I have about 75 vulnerability scans that occurred on 29/06/2026 which is just two days after the resets. After having them run for a day, my token usage just spiked up insanely..
When I checked my token usage breakdown, it suggested to me that the majority of vulnerability scans were using opus models by default when they get run? Am I correct in reading this?
I suspect it is because when I work my other personal repos on cursor mobile like 10 chats, I never chose Opus by default. I alway pick auto for the most balancing type of agentic work - which is generally Composer 2.5 all the time.
Hey, let’s go step by step, because two separate things are getting mixed up here.
Why it says Total 25% but also 80% used
You have two different pools:
Auto + Composer, a bigger included allowance (your screenshot shows 9% used)
API or included premium usage (about 20 premium in Pro), a small pool that premium models charge against like Opus, GPT-5.5, Codex, and especially Max Mode
The Total 25% ring blends both pools, so it looks fine. But your API pool is burned to 81%, and that’s the 80% you’re seeing.
Bugbot is not the main cost, but it’s not free either
In your breakdown, github_bugbot is its own line (11,3%), and claude-4.6-opus (50,6%) is a different line. So the Opus usage is not from Bugbot, it’s from your Opus agent (see point 3). That means the idea that Bugbot uses Opus by default is not supported here, these are separate items. Also keep in mind Bugbot cost depends on effort level, PR size, and how often it runs, so on higher settings each run can cost noticeably more.
What is actually burning the pool
Look at claude-4.6-opus-high-thinking = 50,6% (13,1m tokens). That’s your scheduled Vulnerability Scanner (the Agents dashboard shows about 243 runs on June 29). It’s configured to use Opus 4.6, and likely Max Mode too, which is the most expensive combo, multiplied by hundreds of runs. That’s half your API pool right there.
One important note: cloud or background agents and automations use the model that was selected when you set them up. They have their own model picker, separate from Auto in the IDE. They don’t silently switch to Opus, it just means Opus was picked at setup time.
How to fix it:
Open the Vulnerability Scanner config and set the model to Auto or Composer 2.5, and turn off Max Mode
Also check Max Mode on mobile and web, premium models showed up there too like gpt-5.5-high
This should cut your API pool usage a lot. Opus in Max Mode on automation with hundreds of runs is the fastest way to burn it. Let me know if the numbers don’t change after you switch the model.
Ah I see. Thanks for the comprehensive explanation behind these surging token costs. That is most helpful as well as insightful.
Appreciate the tips and tricks of saving your tokens from burning too much. I’ll bear that in mind going forth.
That being said that, this gives me an idea about building my side projects when working with Cursor AI across my personal github repos.
With the Cursor APIs you listed here, do you have restful endpoint where I can inspect the amount of token data usage at github or workspace repository level over time? Have you catered such an endpoint for something like that?
This would be helpful as I want to monitor the ongoing token data usage across my repos over in my own personal dashboard that I want to build.
On the API question, in short: there isn’t a REST endpoint right now that returns token usage split by GitHub repo or workspace for individual Pro accounts.
A few details to make the full picture clear:
Programmatic usage data is only available via the Admin API /teams/filtered-usage-events, /teams/daily-usage-data and the Analytics API. These are Team and Enterprise features and require a team admin API key. Those endpoints aren’t available on Pro solo. Docs: Admin API | Cursor Docs and Analytics | Cursor Docs
Even there, usage is attributed by user, model, and day, not by a specific repo or workspace. A per-repo breakdown isn’t available on any plan yet.
The API you found at api.cursor.com/v1/… is the Cloud Agents API for managing agents. It doesn’t provide usage or token breakdown.
So for a personal repo-level dashboard, there isn’t a ready data source yet. On Pro, you can only view usage in the Dashboard.
This is exactly what several users have already requested. If you want, you can upvote or add your use case in these threads to help give it more signal:
Thanks @deanrie
Thank for sharing those threads. I’ve given my upvote for both of them and responded with my strong desire to have this feature prioritised as well.
I think it’s going to have major benefit for everyone, if not just myself in the immediate term as quite often, I would like to get immediate feedback how I’m trailing my token usage across all the repos I own or other client owns with my own Cursor baseline Pro subscriptions.
By the end, I need to make sure I’m tracking and managing my token costs well - not to overspend than it needs to be.
Glad we sorted it out. Votes in those threads really help. The more signal there is, the clearer it is to the team that per-repo and per-workspace tracking is in demand.
For now, that option isn’t available, so for cost control keep the key point from that thread in mind: premium models Opus, GPT-5.5, and Codex, and especially Max Mode, can burn through a small API pool fast. Also, automations and background or cloud agents use whatever model you picked when you set them up. So if you want to minimize burn on side projects, use Auto or Composer 2.5 without Max Mode for routine runs.
If you still have questions about the numbers after switching the model in Vulnerability Scanner, ping me and we’ll take a look.
