Hi,
What’s the difference between ‘grok-4.5-fast-xhigh’ and ‘cursor-grok-4.5-high-fast’?
Why was ‘grok-4.5-fast-xhigh’ removed?
Thanks,
Kevin
Comparison of token usage and speed between the two models:
- Tokens: cursor-grok-4.5-high-fast uses twice as many tokens as ‘grok-4.5-fast-xhigh’.
- Speed: cursor-grok-4.5-high-fast is much slower than ‘grok-4.5-fast-xhigh’.."
RequestId: f9ec3683-4153-4db0-abda-83e101d2d6f9
Yes, me too.
After grok-4.5-xhigh was renamed/migrated to cursor-grok-4.5-high, the same underlying Grok 4.5 model consumes drastically more of my included usage — even though it used FEWER tokens. I did not enable fast mode.
From my Included Usage page, same billing period, First-party models:
- grok-4.5-xhigh: 21.0M tokens → 4.6% of quota
- cursor-grok-4.5-high: 16.4M tokens → 6.7% of quota
So cursor-grok-4.5-high used about 22% FEWER tokens but consumed about 46% MORE quota. That works out to roughly 1.85x higher cost per token, for what should be the same Grok 4.5 model at the same $2/$6 list price — and this is happening during the first-week double-usage promo, which should make it cheaper, not more expensive.
This clearly does not match the published pricing. Please clarify:
- Why does the same Grok 4.5 model bill so differently after the rename?
- Is prompt caching still active for cursor-grok-4.5-high? These numbers suggest it is not.
- Is this a billing regression introduced by the migration, and can the previous behavior be restored?
Hey everyone!
Yesterday, we renamed some slugs to match what you see in the UI (High effort in the UI for Grok 4.5 was previously backed by the slug grok-4.5-xhigh, and is now cursor-grok-4.5-high). To be clear: this is purely a naming change he the there should be no impact to billing or performance.
Keep in mind that because token types (Input, Cache Read, and Output) bill at different rates, it’s not possible to directly compare a total token count across models.
Below is how the slugs changed.
grok-4.5-medium→cursor-grok-4.5-lowgrok-4.5-fast-medium→cursor-grok-4.5-low-fastgrok-4.5-high→cursor-grok-4.5-mediumgrok-4.5-fast-high→cursor-grok-4.5-medium-fastgrok-4.5-xhigh→cursor-grok-4.5-highgrok-4.5-fast-xhigh→cursor-grok-4.5-high-fast
Hi Colin,
Thank you for the explanation. I understand that different token types incur different costs. However, for the exact same type of task within the same project, the cost should not vary so drastically. It is currently nearly double the original amount, which is unacceptable. Could you please check if the 50% discount was properly applied to my account?
Thanks,
Kevin
By the way, my previous account was on the Pro+ plan, where 6 million tokens consumed nearly 1% of my quota. I have now upgraded to Ultra, but 8 million tokens still consumed nearly 1% of the quota. Since Ultra is supposed to provide 20x the Pro quota while Pro+ only provides 3x, this math seems incorrect. Could you please look into this calculation error?
ultra:
Looks like there’s no 50% discount for cursor-grok-4.5-high-fast, it’s charged per list API price $4/$1/$18 input/cache/output (see my pre message)
FYI: @Colin
Hey all.
Thanks again for the report. It looks like indeed, when the model slugs change the promotional campaign did not transfer to the new slugs.
We are working through what we need to do to make this right for requests that occurred between the model slugs changing, and the fix.
Hi Colin,
Any updates on this?
We’re still in the process of issuing corrections! Thanks for your patience.
Hi Colin, thank you for the update. Could you please clarify whether the tokens/credits charged without the 50% discount will be automatically refunded or adjusted? Thanks!
Credits have been issued, and while the exact figure isn’t available in the UI (something to improve in the future!), you should now see a number of requests marked “free” under https://cursor.com/dashboard/usage.
“How can I find out the rules for these credits, especially considering the abnormal usage over the past few days?”
Hi Colin,
Your proposal is unacceptable.
- What are the specific rules for distributing the free tokens?
- There was absolutely no prior notice regarding this distribution. We inquired in the community but received no response, which led us to believe it was compensation for the ‘Investigating service degradation’ incident.
