Dear Cursor Team,
I’m currently using Cursor in agent mode with auto-select enabled. This setup is necessary because, otherwise, each of the up-to 25 calls is billed individually — a significant cost issue, which I only encountered after updating to cursor 48.x or higher (Bug #1).
To manage this, I’ve limited the available models in the settings to only those that perform reliably. However, even with these restrictions, Cursor still selects GPT-4/4.1 in auto-select mode, despite it being disabled in the settings (Bug #2).
This is highly problematic, as GPT-4.1 consistently delivers poor results for coding tasks, leading to wasted time and increased costs. At the moment, this severely impacts the usability of Cursor for my workflow.
I would greatly appreciate your assistance in resolving these issues.