The Frustration of Not Being Able to Set Per-Window Models in Cursor

I’ve been using Cursor for a while now — and overall, it’s become an integral part of how I code and experiment with different AI models. But there’s one thing that’s driving me absolutely nuts: the fact that I can’t set the model per window.

Let me explain.

Cursor lets you pick which model you want to use — GPT-4, GPT-4 Turbo, GPT-4o, or whatever else you have access to — but that choice is global. So if I have multiple windows or projects open, they’re all forced to share the same model setting.

This doesn’t make sense for how I work.

When I’m hacking away on a quick prototype or testing throwaway ideas, I’m happy to use a cheaper or faster model. It doesn’t need to be the top-tier option — speed and cost matter more in that context.

But in a serious project window — something I’m actually shipping — I want the best reasoning and code quality the model can offer. That’s when I want GPT-4 Turbo or GPT-4o turned on and ready.

Right now, that workflow is impossible. If I switch the model to save tokens in my throwaway scratchpad, my main project window gets downgraded too. And vice versa — if I switch my main project to GPT-4o, my scratchpad is burning through expensive tokens just for quick tests.

It’s a mess.

It feels like a tiny limitation, but it breaks the flow completely. Instead of Cursor adapting to how I work, I have to adapt to how Cursor wants me to work. I find myself constantly second-guessing which window I’m in, switching models back and forth, or wasting usage on things that don’t need high-end reasoning at all.

I know I’m not alone — there’s already a thread on the Cursor forum where other users are asking for this exact feature. The thread is growing, but so far, it’s just radio silence from the devs.

So here’s my plea: if you’re building a tool for devs who use AI as an assistant across multiple contexts, please let us choose a model per window or project. It’s such a simple tweak that would massively improve the workflow for anyone juggling experiments, side notes, and production code at the same time.

Cursor is fantastic — but this one missing piece makes it feel unnecessarily rigid. I hope they fix it soon, because I really don’t want to switch tools just to get back the flexibility I need.

If you’re frustrated by this too, join the forum thread and let the team know. The more voices, the better. Maybe they’ll finally bump it up the roadmap.

Until then, I’ll be here — manually switching models back and forth, trying not to burn through tokens for no good reason.