Hello. Does Cursor currently support the “Pro” model of GPT-5? I did not see it as a separate model selection, so I am thinking “no”, but wanted to confirm. And if indeed the answer is no, are there plans to support in the future, and if yes, any idea of timeline?
(In case there is any ambiguity of what I am referring to, it is discussed in some detail here GPT-5 vs o3 vs 4o vs GPT-5 Pro — 2025 Benchmarks & Best Uses and briefly mentioned here https://openai.com/index/introducing-gpt-5/ )
GPT-5 Pro would be great.
If you didn’t already know:-
based on what OpenAI has disclosed, “Pro” isn’t a totally different neural network.
It’s the same underlying GPT-5 model weights, but run in a different execution mode:
Higher compute budget per request – it spins up more parallel chains of reasoning or
uses a longer “thinking” phase before writing the answer.Different system-side inference settings – more tokens for reasoning, deeper search through possibilities, possibly enabling more verification passes before output.
Routing logic – the request router decides whether to invoke gpt-5-thinking-pro based on complexity, or you explicitly get it because you’re a Pro tier user.
This is similar to how “gpt-4-turbo” and “gpt-4” in the past shared weights but ran with different latency/price trade-offs — except here the difference is aimed at more accurate and thorough reasoning, not just speed/cost.
So: same brain, but given more time and mental “scratchpad” space before answering.
so you could more-or-less get the same output, by just telling it in advance to think more, or making it do 2 passes at everything (think first, act in a second pass).
so it’s similar to the current ’gpt-5-high’ model?
GPT-5-Pro is not released in the API yet: https://platform.openai.com/docs/models
If I read this right:-
https://openai.com/index/gpt-5-system-card/
GPT-5 is not “A model”, it’s a router, which picks a model based on a “smart guess” at the start as to which one you need (or in other words - can they get away with giving you a sketchy answer by using a cheaper–in-electricity-to-them model’s answer)
isn’t it the same as high reasoning gpt-5 in Cursor?
Thanks to everyone for their input and thoughts.
This has been made available by openai and i see it in cursor models as max only, anybody tried it yet? according to open ai graphs, normal gpt 5 is 87% and gpt 5 pro is 89%
Can anyone share their personal experience on whether GPT-5 Pro is worth using? I noticed it has been added to the available models.