2026年9月18日 8:21 ET 发布 · 更新于 2 个月前 · 1 个来源
Maxing reasoning effort on GPT-6 Astra wastes compute for no quality gain
X
·
原始报道
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.
原文