maybe the openai moat is the data we shared along the way.
no seriously, openai seemingly lost interest in being the 'best' model - instead optimizing for other traits such as speech and general human likeness? there's obviously codex but from my experience it's slower and worse than the other big 2 in every single way: cost, speed and accuracy. codex does seem to be loved by vibe coders the most that don't really know how to code at all so maybe it is also what they're optimizing for and why it doesn't personally suit me.
others might have better models, but openai has the users emotionally attached to the models at this point even if they know it or don't. there were several times I recommended switching and the response I got is that "chatgpt knows me better".
haven't done bug hunting with openai so cannot comment, but given the right tools opus can most definitely solve bugs as well.
I think opus is special because it was trained explicitely on very strong reliance on such tools while gpt seems to "know" how things work a lot better reducing the need for tools.
opus being able to dedicate a lot more parameters for these things make it a better model if you give it what it needs, but that's just my observation. it's also much faster as a bonus.
no seriously, openai seemingly lost interest in being the 'best' model - instead optimizing for other traits such as speech and general human likeness? there's obviously codex but from my experience it's slower and worse than the other big 2 in every single way: cost, speed and accuracy. codex does seem to be loved by vibe coders the most that don't really know how to code at all so maybe it is also what they're optimizing for and why it doesn't personally suit me.
others might have better models, but openai has the users emotionally attached to the models at this point even if they know it or don't. there were several times I recommended switching and the response I got is that "chatgpt knows me better".