Turns out it’s confirmed that Company A’s model started getting dumber since the 4th of last month because default reasoning was adjusted from high to medium. Also, I just checked, this bug should’ve already been fixed, but they’re just holding it back, waiting to release it with GPT 5.5…