gpt-6IMAGE

Setting Astra's thinking depth to medium in Codex proves smarter and cheaper, user finds

A user reports that medium thinking depth for GPT-6 Astra in Codex is more efficient and smarter than the highest setting, drains quota more slowly, and works without spawning excessive agents.

冰河

@bingheArchived 2026.09.14Originally posted 2026.09.08

Setting Astra's thinking depth to medium in Codex proves smarter and cheaper, user finds

冰河 @bingheOriginally posted 2026.09.08

Open original on X

我把 Codex 的 GPT-6 Astra 思考深度调到“中”，突然感觉额度消耗也不快了，而且还变聪明了不少。。竟然比最高思考深度做的更好。。这真是奇怪了。实测比 Sol 的最高做的更高效，没有开非常离谱多的智能体。。简单直接，还高效！

x.com/binghe/status/2097224697343836599Archived 2026.09.14 16:18