Date: September 14th, 2026 8:00 PM
Author: https://i.imgur.com/YbMiOBE.png
1. they disabled thinking traces, allegedly to prevent distillation
2. but anyone who's seen the thinking traces sees the problem: loop after loop after loop. The thinking traces keep saying "wait, the user said..." It's your original prompt being read -from scratch - by another layer in the model. For some reason they decided this was the way. Loop after loop after loop until you get perfection. That's why one prompt costs $40 to process now
(http://www.autoadmit.com/thread.php?thread_id=5903948&forum_id=2/en-en#50136020)