Cursor Token Rate seems to have been designed to push people to use more Composer or Grok 4.7 models instead of keeping genuine promise of allowing access to other models at API rate (considering Cursor already probably gets discounted access to model API rates because of their volume).
With Cursor Token Rate of 0.25$/MTok, accessing Luna (which is better than composer 2.5) 100%+ more expensive compared to using it via other tools. For cached tokens, CTR makes Luna model access 1000% more expensive.
There is not much harness difference between Cursor and other competing tools. It will be worth if cursor considers reducing CTR to be 10% of model’s input rate with max of 0.25$ per million token or some such scheme that does not monetarily punish users (or fleece them) for using other cheaper alternative models.
Will it be possible to reduce or wave off CTR on Luna?
Hi @oknrd Thank you for the post. Enterprise contracts can be specialized, so I don’t want to respond with general guidance before consulting your team’s account representatives here at Cursor. We’ll get back to you shortly (possibly over email). Thanks for your patience.
Ohh, now I realize what happened . I see today morning that we got some response. So I will let them handle this.
Neverthless, this question was intended for my personal pro account perspective and I didn’t realize that I have logged in with cursor with different account. You can treat this as non-enterprise ordinary pro-account question.
Got it! For individual plans like your Pro Plus plan, there is no Cursor Token Rate on any model. For enterprise and teams plans, there is no Cursor Token Rate on any Cursor models. Lmk any follow-up questions!
They don’t get negotiated. So the $0.25/million on cached input tokens for luna absolutely blows up the cost and makes it not worth it at all to use Cursor if I want to use the GPT-5.6-luna model
On Teams, third-party models (including GPT-5.6 Luna) include a Cursor Token Rate of $0.25 per million tokens. That rate applies to input, output, and cached tokens, on top of the model’s API price. Cursor models such as Composer 2.5, Grok 4.5, and Grok 4.6, plus Auto Cost, do not include it.
If you want to avoid the rate on Teams, use Auto Cost or a Cursor model. More detail is here: Cursor Token Rate | Cursor Docs
I have the same issue where I’m being charged 7.5x luna’s API cost. A percent based fee would make a lot more sense IMO than just charging $0.25 per million tokens on a model that charges $0.02 per million cache read. It’s practically making luna a more expensive model to run than sol, as smaller models tend to use more tokens to think.
. I see today morning that we got some response. So I will let them handle this.