Since all new models are now Max by default, it’s frustrating that trying models like GPT-5.4 or Opus 4.7 eats into the 500-request plan.
It would be really great to have a toggle between API pricing and the request-based plan, so users can try newer models without draining their request balance.
As it stands, I have to burn through my requests before I can even test them.
I’m not asking to get anything for free. I’m literally asking for the ability to be able to pay API pricing for the new models
Cursor has already moved to a usage-based pricing model that works pretty much the way you’re describing. Instead of the old 500-request system, there are now two separate usage pools:
Auto + Composer pool for everyday tasks with fixed token rates
API pool when you pick a specific model like GPT-5.4 or Claude 4.7 Opus, you’re charged at that model’s actual API rate
The Pro plan costs $20/mo and includes $20 of API usage per month. Once that’s used up, you can enable on-demand usage in your dashboard settings https://cursor.com/dashboard/billing to keep going at the same API rates.
If you’re still on the old 500-request plan, you can switch to the new usage-based system from your billing dashboard https://cursor.com/dashboard/billing.
Ah, got it. You want to keep the 500-request plan, but pay API rates separately for frontier models. Unfortunately, that kind of hybrid setup isn’t available right now.
One important detail though. The new Pro plan with pay-as-you-go billing (20 USD/mo) splits your budget into two separate pools:
Auto + Composer pool for everyday work with lower token rates
API pool (20 USD included) for frontier models like Opus 4.7 and GPT-5.4 at real API rates
The key difference is that frontier models don’t use your everyday pool. On the old plan, everything comes from the same 500-request pool. The exact cost per request depends on your usage. You can check pricing here: Models & Pricing | Cursor Docs.
Pretty please! Just give us a little toggle in the settings that swaps us on to API pricing while keeping our 500 intact. I imagine it would be so easy to implement.