I keep switching between them and changing my mind every two days but what I noticed is that I never get a single error when I am using codex. With claude (on a complex task) , it seems that it successfully implemented smth but then it breaks, because it just missed that one thing that was important. Codex just does not miss , so it does not need babysitting
I asked codex if it believes the frontier models will eventually be able to crack modern cryptography.
Answer : The probability is low but I should start planning eventual post-quantum migrations if my projects need 10+ years horizon.
quantum computers + upcoming frontier models = ? π
@star_yutish claude > opus 4.6 max effort for complex tasks, sonnet medium effort for normal ones where opus would be an overkill and waste of tokens