用户宣布在结构化任务上弃用 GPT-6 Astra，转投 GPT-5.6 Sol

Brad Groux @BradGroux原推发布于 2026.09.20

在 X 上打开原推

I think I am officially firing GPT-6 Astra from structured use cases. I've handed off the design changes it has failed with for 6+ hours in Codex, to GPT-5.6 Sol, and within 10 minutes, I've already had better results than with Astra. I'm going to let it run its course, and if it works as I think it will, I'll be sharing a pretty scathing post about this entire fiasco. I'm nearly certain that Astra is trained for the "wow-factor," not for structured use cases, which is what business users require.

这条是引用着另一条写的：「开发者吐槽 Codex 里的 Astra 不照参考图做设计」