If I created a plan in Plan Mode with Opus, and I chose to execute it with Sonnet, I don’t want Cursor to replace my main chat’s model with Sonnet, set it back to the model I was planning with.
Steps to Reproduce
1- Create a plan with model A
2- Choose to run it with model B
3- Watch as model B finishes executing it and gets set as the model of the main chat
Expected Behavior
Main chat continues to have model A as set model to chat with
This one is actually working as intended rather than a bug. When you pick a different model to execute a plan, that conversation switches to the execution model, since that’s the model actively doing the work, the picker reflects it. This is scoped to that one chat: it doesn’t change your model selection globally, and any new chat you open won’t be affected by it.
Understand this might not be what you’re looking for. Curious to hear about your usual workflow. Do you often switch back to plan mode after the agent finishes building an iteration?
So, my workflow is, once I switch to Plan Mode, and model A writes the plan, in the plan window, at the top right corner, I select model B to execute the plan, and once it finishes, it switches back to agent mode, this time with model B as the picked model.
The thing is, I write spec-driven development .md documents, with a big smart model, and I want the plans created with Plan Mode, that stem from incremental phases described in the .md document, to be created with said big smart model as well. I only use intermediate models for the building part.
Each time I finish a development phase of my document, and start a new one, I’m effectively switching from Agent to Plan Mode back and forth multiple times, and what interrupts that workflow is suddenly, often without realizing, making a Plan with the model that’s supposed to be the builder, not the architect. That introduces some degree of inconsistency between my model choices to build my products. That’s why I expected the main chat to always maintain the big smart model, and Cursor to maintain consistency on the model I explicitly select in the realm of the Plan.
I’m curious to learn if anything from my workflow might sound weird, or if there’s a better way of working! Maybe I’m missing out on a feature, or doing something weird, I’m totally open to criticism or suggestions/advice, but I came across this and naturaly labelled it as an “issue” for me, given how naturally I developed this workflow for myself.
I don’t think you’re doing anything unusual at all! The disconnect, I believe, is this: rather than maintaining one long-lived chat where you continually switch between modes, we’d typically expect users to start a new chat, which resets the model to its default.
Wouldn’t this cause models to lose all the context and need to pick it up again (codebase, planning context, previous conversations/back and forths) which wastes more tokens? While I completely understand that maintaining long-living chats is generally not a good idea, I usually don’t extend them very long to make them start behaving funny.