I’m trying to understand how the Cursor Models included usage percentage is calculated for Pro users.
I compared my usage with a friend who is also on Pro. We both mainly use Grok 4.6 for similar software development / coding workloads.
My usage
Current billing cycle usage:
87.1M total tokens
Cursor Models: 100% used
Other Models: 25% used
Mostly Grok 4.6 high / xhigh
Very little Fast usage
My friend’s usage
Current billing cycle usage:
303.9M total tokens
Cursor Models: about 65% used
Also mainly Grok 4.6
Similar coding workload
So my friend has used roughly 3.5× more raw tokens, while my Cursor Models usage has already reached 100% and additional usage has started consuming the Other Models pool.
I understand that raw token count is not a direct measure of included usage because the effective cost can differ depending on:
input tokens
cached input
output / reasoning tokens
effort level
Fast mode
context reuse
However, since both of us mainly use the same model for similar coding tasks, the difference still seems unusually large.
My Cursor usage is pretty high too. I can usually get around 400–500 million tokens with Grok 4.6 before reaching 100% usage, so your situation definitely seems abnormal.
Hey, thanks for the detailed breakdown and export, it helps.
The main reason you and your friend see different numbers is the plan, not how you code. Your account is on a student promo plan, and your friend has a smaller included Cursor Models allowance than standard Pro. So the percentages between your account and your friend’s account aren’t directly comparable, even with the same model and similar tasks.
A few notes on metering so it lines up:
Included usage is based on request compute cost, not raw token count. Cache read tokens are much cheaper than fresh input, and output and reasoning are more expensive, so two accounts with similar token counts can burn the pool at very different rates.
Fast variants cost more than the regular ones. Your export shows a bit of usage on Fast Grok variants, and that also adds up even if it felt like Fast was barely used.
Once the Cursor Models allowance is used up, more usage on those models starts billing from the Other Models pool. That’s why you see 25% on Other Models even though you barely use third-party models directly.
The best way to verify the real cost per request is the usage page in the dashboard: cursor.com/dashboard/usage. It shows the cost of each request and it’s more accurate than comparing total token counts.
Let me know if anything still doesn’t add up after that.
Hi thanks for the reply,
Just to clarify one point: did you mean that my student promo plan has a smaller included Cursor Models allowance than standard Pro?
Hey, sorry for the confusion in my wording above. I phrased it poorly.
Basically, the included allowance how much Cursor Models usage is included in the plan is tied to the plan, not to how you code. So your percentages and your friend’s might not be directly comparable, even if you use the same model and have similar tasks.
I don’t want to give you an inaccurate number for the allowance on the student promo. I’ll double-check the exact included limits for that plan so the comparison is fair, and I’ll come back here with an answer.
What does definitely work this way for the usage metric:
Included usage is calculated by the request’s compute cost, not the raw token count. Cache read tokens are much cheaper than fresh input, and output and reasoning are more expensive, so two accounts with a similar token count can burn through the pool at very different speeds.
Fast variants cost more than the regular ones. Your export shows a bit of usage on Fast Grok, and that adds up too.
Once your Cursor Models allowance is used up, further usage of those models starts being charged from the Other Models pool. That’s why you see 25% on Other Models even though you barely use third-party models.
The most accurate per-request cost is shown at cursor.com/dashboard/usage. Let me know if anything still doesn’t add up after I confirm the limits.
