Tried Claude 3.5 Sonnet against GPT-4o for coding tasks for one week

I spent last week testing both Claude 3.5 Sonnet and GPT-4o on the same Python projects at work. Claude nailed a complex data parsing script in one shot that GPT took three tries to get right. The difference was huge on debugging too - Claude found a nested loop error in about 20 seconds where GPT kept suggesting unrelated fixes. Has anyone else noticed this gap widening, or was my test just too small?

2 comments

2 Comments

sam_wood601mo ago

Claude nailed a Python refactor for me in one try last month that GPT had previously choked on three times. The gap on debugging complex stuff has definitely gotten wider in my experience. Your test sounds about right based on what I've seen.

the_jason1mo ago

Same thing happened to me with a regex problem last week. Claude fixed my garbage pattern in one shot while GPT kept adding more broken logic to it. My dumb mistake was using a greedy quantifier where I needed lazy matching, and Claude spotted it instantly. Pretty embarrassing when an AI has to tell you your regex game is weak lol.