does anyone else notice that claude opus is smarter at lower traffic times (i.e. right now, sunday afternoon)? this model is good, but the inference policy is so weird