Published Sep 18, 2026, 8:21 AM ET · Updated 2 months ago · 1 source
Maxing reasoning effort on GPT-6 Astra wastes compute for no quality gain
X
·
Original Report
Running GPT-6 Astra at maximum reasoning effort is the wrong approach because low-effort Astra already beats the previous model at high effort, so maxing the slider just pays twice for the same answer.