gave opus my repo, told it to loop for 5 hours and use gpt sol for the boring stuff, went outside came back to 94% -> 98% on my evals, replies 45% faster, tts cost down 38%. Opus is unbelievable
I think I'm switching back to claude after more than a half year with codex. The new opus 5.5 is the best model i've ever used. I Hope limits stay this generous.