Less errors. Better answers. More effective answers. More thinking. Better problem solving.
GPT-5 spends minutes solving an issue and it actually solves with a good solution and I only got 1 error with it. Just a single error.
Claude 4 sonnet is fast, but gives out a lot of errors and sometimes I need to spend 5 prompts to solve something and it does not solve it.
Agree.
I am honestly not understading why is it getting so much hate.
It gets hate because it is a routing model. It routes your queries to the most relevant model, according to complexity, avail resources, usage, etc.
Initially one prompt, problem solved. Now few prompts. I feel quality has decreased because many are using it. Possible, especially with how Cursor work?
GPT-5 is way better than Claude 4 Sonnet/Opus … At least it follows all the rules , prompt and memories for real ! And the code it produces is really high quality too … at least for Rust , Typescript and Javascript ! I love it a lot 
Could be.
Here, the quality is the same. Pretty good still.
Waiting for Claude 5 now to break GPT-5. But for now, I am going with gpt.
Remember to come back here and post again when you realise the claude4→gpt5 switch turned out to be a mistake - there’s a HUGE number of unhappy people (me included), so I think anyone finding “gpt5 better” either isn’t doing serious code, or just got lucky.
It depends on what you are doing, I always keep going back to claude because of the way it can use tools effectively and also the way it retains context. I’m sure it’s just a matter of tweaking tht GPT settings to get it to use tools more as they did with gemini. Gemini had these problems in cursor in the beginning also..
But even after cursor fixed almost all the issues with the other models like gemini/grok etc , Claude still in my opinion always uses tools better to figure things out , although it’s over-eager in the responses so some temperature tuning would be nice.
GPT5 is definitely good, i’ve tested every single model that cursor has and like i said before i always end up going back to claude but gemini is getting closer, GPT looks like it could really work very well once the intergration is better (tools, rules, mcp servers all that)
in general, is a very good model. I think the problem is just cursor is broken using GPT5.
this
GPT5 tackles problems differently for sure, seeing missing links that Claude didn’t. It seems like it also does better at recognizing and telling you how different types and systems are related, compared to Claude. Whether it’s right or not is a different story, but it verbally tries to distinguish things to you.
I experience GPT5 blatantly ignoring my instructions and running off way too much. Biting off more than it can chew and smashing many lines of thoughts into a single sequence. It’s like it believes it’s an expert programmer and tries to do things it’s own way instead of listening to me.
Other users are saying it’s just a matter of Cursor team fine-tuning the model, so I am hopeful for the near future. But for now, I’m sure I will go back to Claude after the free period is over, and just use GPT5 once in a blue moon for it’s unique second opinions.