Codex 的子 Agent 调度有个问题—或者更准确地说,是 ChatGPT 的问题。
spawn_agent工具不能选择模型,也不能设置推理强度。
所以,每次 GPT-5.6 Sol Ultra 创建一个子 Agent,跑起来的还是另一个 Sol Ultra 实例。
这就是为什么你的额度掉得这么快。
There's a flaw in Codex's (or should I say ChatGPT's?) subagent orchestration.
The spawn_agent tool doesn't let you choose the model or reasoning effort.
Therefore, every time 5.6 Sol Ultra spawns a subagent, you're getting another Sol Ultra instance.
That's why your quota gets drained so fast.
显示更多