Hi,
I’m wondering how the pool works now that it has Auto + Composer. Auto is priced higher than Composer 2 so it will eat up usage faster, right?
Hi @thddrew great question.
Auto has a slightly higher per-token pricing schedule than Composer 2 regular. However, there’s some nuance here, because Auto doesn’t always route to Composer 2. Auto selects from models that balance intelligence, cost efficiency, and reliability. In some cases, Auto will route to more expensive models (yet you still get the flat rate of Auto).
To answer your question directly, if you set your model picker to Composer 2 regular, that will consume your usage more slowly than Auto.
Check out the full pricing here: Models & Pricing | Cursor Docs and let me know if you have any questions!
The wording is not clear for me. What do you mean by auto will get the same flat rate but it will consume usage faster than Composer 2 ?
Hi, sorry for any confusion. Auto will get the same flat rate regardless of the underlying model used under the hood. So for example, let’s say Auto routes your request at some point to a frontier model. You’re not billed at the frontier model rate, you’re billed at the Auto rate. It’s kind of like at a rental car place, you book a sedan, but the sedan’s not available, so they offer you a convertible sports car instead for the same price you paid for the sedan.
You can see the exact pricing here: Models & Pricing | Cursor Docs
Composer 2 Regular is slightly cheaper than Auto on both output and input tokens. Note that Composer 2 Fast Mode is slightly more expensive than Auto.
Thanks Kevin. I know we won’t know which model “Auto” will route to, but are the latest frontier models included in the pool? eg. opus 4.6, 4.7 or gpt 5.4, 5.5.
Auto can route to frontier models, but it does not guarantee a specific model/version for a given request. The Auto pool is dynamic and optimized around quality, speed, reliability, and cost, so the exact model may change over time.
If you specifically want Opus 4.6/4.7, GPT-5.4, or another named model, the best path is to select that model directly from the model picker, where available.
Thanks for the answer.
When I disable models in cursor settings, does it affect the model pool for auto or it is just a GUI change ?
Disabling models in Cursor Settings is a client-side preference. It hides those models from your model picker, but it does not change the Auto pool on the routing backend. So Auto can still route to a model you’ve toggled off in Settings.
If you want to be sure a particular model is used (or avoided) for a given request, the most reliable approach is to pick a specific model from the model picker instead of using Auto.
(Note for completeness: team admins can restrict models from the team dashboard, and that restriction is enforced server-side — but that’s a team-level control, not the personal toggle in Cursor Settings.)