Claude 3.5 vs GPT-4o Coding Test Comparison Review
I tried a few Baekjoon Gold-level problems, and Claude performed better than expected.
1. GPT-4o: Clean time complexity explanations, but frequent implementation errors.
2. Claude 3.5: High code completeness and provides debugging tips.
Personally, Claude seems better for algorithm problems. But I haven't tested fine-tuned models yet.
10 answers
Agreed, Claude was more accurate on the gold-level problems for me too.
Oh, this is right lol. Giving debugging tips is actually way more helpful than I thought.
GPT does make mistakes in implementation, lol. It gets things wrong surprisingly often.
Well, I just find GPT more convenient. Claude's explanations are too verbose and I can't focus.
Claude seems to have really improved. Its coding used to be a mess.
But fine-tuned GPT-4o might actually be better, right? Comparing it to the general model seems a bit unfair.
I also tested the platinum-level problems, and Claude had fewer timeouts. However, GPT was better in terms of memory usage.
Honestly, GPT seems better for beginners. Claude is too focused on experts.
GitHub Copilot is still the best lol, especially for Python.
Please let me know the source. Did you run it yourself, or did you refer to another benchmark?