↑
jdw64 3 hours ago
In my case, I found that Claude is better for frontend design and initial structure, while GPT tends to perform better on core logic.
Claude often misses basic security concerns — not even OWASP-level issues, but simple things like proper XSS headers. GPT, on the other hand, tends to overcompensate with too many defensive checks, sometimes introducing unnecessary complexity.
That said, for domain logic where failure is not acceptable, I’d rather have too many safeguards than too few. In those cases, I’ve had a better experience with Codex than with Opus.
However, Claude is much faster and better at generating a solid skeleton. Personally, I struggle a lot with starting from a blank page. I can modify and refine code, but writing from scratch tends to overwhelm me. So my workflow is to let Claude generate the initial structure, then I refine it, and finally I use GPT as a strict reviewer.
GPT works well as a validator. It tends to consider far more edge cases than I would. But when it comes to writing documentation, Claude is clearly better. Since I rely on documentation to remember what I’ve built, that matters a lot to me. In general, Claude is much better at summarizing and explaining things in a way that others can understand.
Another issue I’ve run into is encoding. Codex models frequently break UTF encoding, which is a serious problem for me. I often work with non-English environments, and especially with XAML projects or code that needs to be delivered to Chinese clients, encoding issues are critical. Codex has been quite weak in this area.
Separately, Opus provides a much better overall developer experience than Codex. Codex feels heavier as an application and tends to freeze more often. While Codex can produce better code at the function or method level, Claude fits my overall workflow better.
In the end, I think the “best model” depends heavily on how you work. For my workflow, Claude works better overall, even if Codex sometimes produces more precise code in isolation.