Feedback on Uncontrolled composer-2-fast Usage in Cursor Plan Mode (Legacy 500 Monthly Request Subscriber)

Ideas Feedback
,
1

Dear Cursor Team,

I’m a long-time legacy subscriber with a 500 monthly request limit (not the new token-based billing plan), writing to share a key usability issue I’ve encountered in Cursor’s Plan mode that directly impacts the fairness of my quota usage.

When using Plan mode, I explicitly select Opus 4.6 for my coding tasks. However, during the interactive Q&A phase of Plan mode—where the tool asks follow-up questions to refine the plan—I’ve noticed that every time I submit a response to these questions, the system automatically calls composer-2-fast in the background, consuming 2 request credits per interaction.

This is problematic for two core reasons, especially for legacy per-request subscribers like myself:

  1. Lack of user control: I never opted in to use composer-2-fast for Plan mode interactions, and there is no way for me to disable or prevent this automatic background call. It operates entirely outside of my explicit choices.

  2. Unnecessary double consumption for per-request users: The model used for user dialogue in Plan mode should align with the model I’ve selected (Opus 4.6), rather than defaulting to a model that deducts 2 credits per turn. For legacy subscribers on a fixed 500 monthly request limit (not token-based billing), this creates an unexpected, unmanageable drain on my monthly quota that I cannot mitigate.

As a legacy per-request subscriber, I believe the interactive dialogue steps in Plan mode should use the model the user has explicitly chosen, without introducing an uncontrolled model that incurs double credit consumption. This would ensure a transparent and fair experience for all users, especially those on legacy fixed-request plans.

I hope the team can look into this logic and adjust it for a more user-centric design. Thank you for your attention to this matter.

image
image1009×176 5.21 KB

image
image169×198 3.22 KB

1 Like
3

Thanks for the clear write-up and screenshots.

Plan mode runs as an agent flow. The model you choose in the picker is still the main model for the agent, but additional model work can happen as part of that flow (for example when the agent does broader exploration). Some of that work is served by Cursor Composer models, so your usage history can list composer-2-fast (or similar) even when the picker still shows Opus. That can also show up as more than one request for a single “step” from your perspective, depending on how many model calls ran.

I hear you on legacy fixed-request plans: if the UI highlights one model but the ledger shows another, it feels unfair. There isn’t a separate toggle today to force every sub-step onto the picker model.

For how your plan attributes requests to specific models, use the in-app usage breakdown and our docs: Usage limits and Available models.