When using the Agent sidebar panel in the Cursor IDE, the selected Model reverts to Composer 2 Fast whenever opening a new Chat.
I am using Composer 2 (non-fast), however I keep finding that it reverts to Composer 2 Fast whenever opening a new conversation tab or after a few prompts.
Steps to Reproduce
Open a new conversation, set the Agent to Composer 2, give it a prompt, then open a new tab - it will have reverted to Composer 2 Fast.
Expected Behavior
The Agent should stay on Composer 2 - the latest Agent that I have manually selected.
This is a bug. When you open a new chat, the model selection for Composer 2 is being silently switched to Composer 2 Fast. The team is aware of similar model persistence issues, and Iāve flagged this specific case. Thereās no timeline yet, but your report helps us prioritize.
As a workaround for now, try selecting the model again after opening a new tab. Unfortunately, thereās no permanent client-side fix until this gets patched.
Iām using the new Cursor 2 model, but I want to use Cursor 2 and not Cursor 2 Fast because the former is 1/3rd the cost. Yet it seems that Cursor is regularly flipping me over to Cursor 2 Fast. Itās very frustrating. Is anyone else experiencing this?
Steps to Reproduce
Set in chat to use Composer 2.
Expected Behavior
Stays with Cursor 2 and does not change to Cursor 2 Fast without me selecting it.
For example Composer 2 randomly becomes āfastā (which is 3x more expensive) or GPT 5.4 switches to 1M context.
It just burns my tokens. Idk how much it burnt, but rn im on 50% of limit on my third day after reset. Last month I used up my limit a week before the reset (with pretty same overall usage), soā¦
My default model is composer 2 (not āfastā) but for some reason i often see that cloud agent uses a āfastā one. More over, when i turn off āfastā option by myself - it randomly is turned on again.
How can this be a bug when we have seen it across several of your changes to how models are selected? Like in the past 60 days the UI has changed MULTIPLE times and in each incarnation, fast always gets chosen. Itās a very convenient bug for you. You must understand that your billing practices have not been as transparent as you think ā just look at the user complaints! People are tired of being tricked by dark patterns.
In fact, there are complaints about this exact issue going back to November 2024. But itās ājust a bugā? Cursor Model keeps reverting to fast
So if this isnāt a dark pattern it then implies that your practices are so bad that you keep having the same regression over, and over, and over, for years.
Hey everyone,
The issue with the model reverting to composer-2-fast on each new chat has been addressed in a recent Cursor update. Updating to the latest version should resolve this ā let me know if youāre still running into it!
I was using composer-2.5 through Cursor SDK in an automated tool, but in the billing it is all Composer-2.5-fast. Maybe thatās why it sucked up *all* of my Ultra plan in a couple of hours?
Could it be that the same bug persists through SDK usage?
It is not fixed in latest version, moreover, when you hit rate limit and being asked for switching to Auto, it is changed to composer 2 fast instead of Auto.
The composer/auto pool is very generous.