GPT-5, or back to o3?

Discussions
1

I had some good results in debugging with “GPT-5 high fast” in the beginning - easily solved 1 bug I couldn’t crack using Claude Code Sonnet, or o3 in Cursor.

But not anymore, it’s just slow and dumb now with “GPT-5 high”. Maybe “fast” is the difference, someone mentioned it feels totally different (it should not..)?

I’m leaning towards going back to o3 for debugging, complex planning, and optimizations (all base work with Claude Code).

What do you guys think - GPT-5 vs o3?

1 Like
2

There seems to be a BIG diff between

GPT-5-high and
GPT-5-high-fast

as someone pointed out here. The latter is MUCH better, and FASTER! New price 2x is ok (caching).

Will experiment more with this one.

3 Likes