@kukatCPA 设置思考量,怕降智的看这里 中发帖

看到有讨论CPA反代Codex降智的问题,其实CPA文档里讲的很清楚了 


个人不建议在config.yaml里直接覆盖所有的请求都为xhigh,简单的任务auto/high就够了,
(xhigh是真的慢,最好在high解决不了的时候才换成xhigh)
在客户端自己设置需要的思考级别
这里要注意cherry studio内置的思考级别没有xhigh,可以参考我的做法在CPA里手动添加model
[Screenshot 2026-03-07 at 3.51.25 PM]
完整列表如下
[Screenshot 2026-03-07 at 3.52.09 PM]
opencode 示例
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"myprovider": {
"npm"...