Realworld benchmark between Codex 5.3 and Opus 4.6
swe-agi.comI do feel like the latest codex 5.2 and 5.3 have been really excellent in coding and have been giving opus a good fight. I still prefer Opus 4.6 as my daily driver but specifically for coding tasks I think codex 5.3 is the best, especially when considering value for money.
Another thing I like about codex 5.3 is that its CLI support queueing the message directly without using third party plugins. And it can run weeks without any issues, the CC used to have memory issues and stackoverflows.
This is the benchmark between the latest models on a new programming language to avoid overfitting. Latest models are quite good over generalization to new languages, they can write tens of thousands of lines of code in one prompt that just works.