Quota exhausted: keep official login, switch to relay; Desktop stuck on gpt-reserve, use IDE for now 额度耗尽：保留官方登录切中转；桌面端卡在 gpt-reserve，建议先用 IDE

[szyuing]

English

When ChatGPT Plus ordinary Codex quota is exhausted, Desktop enters Luna Reserve and forces gpt-reserve.

What I want:

• Switch to a third-party API relay
• Do not sign out of the official ChatGPT/Codex account
• Keep CC Switch local routing on

What actually happens on Desktop:

• The official login still shows (expected if official auth is preserved)
• The model picker stays on Luna Reserve
• Requests are sent as model: gpt-reserve
• CC Switch catalog mapping (gpt-reserve → a real relay model) does not rewrite this
• The relay returns 503 No available channel for model gpt-reserve

Current workaround:

1. In CC Switch, enable the relay provider. Keep routing on. Do not log out.
2. Do not continue on Codex Desktop.
3. Open the IDE Codex extension (or CLI).
4. Start a new thread and pick the relay’s real model, not Luna / gpt-reserve.

Same account and same proxy: IDE does not lock to gpt-reserve, and the relay works.

Feature request for CC Switch: when forwarding /responses, rewrite gpt-reserve / gpt-5.6-luna to the mapped upstream model so Desktop can keep official login and still use the relay.

Request for Codex Desktop: while Reserve is active, do not hide custom/routed models, and do not override the request model to gpt-reserve when using a custom base_url.

---

中文

ChatGPT Plus 的 Codex 常规额度用完后，桌面端会进入 Luna Reserve，请求模型被锁成 gpt-reserve。

我想要的是：

• 切到 API 中转站
• 不退出官方登录
• 继续走 CC Switch 本地路由

桌面端实际结果：

• 账号仍显示官方登录（保留官方登录时这是正常的）
• 模型停在 Luna Reserve
• 发出去的仍是 gpt-reserve
• CC Switch 映射表改不了这次请求
• 中转站没有这个模型就会 503

目前的解决办法：

1. CC Switch 启用中转供应商，路由开着，官方登录不要退。
2. 暂时不要用 Codex 桌面端继续聊。
3. 改开 IDE 里的 Codex 插件（或 CLI）。
4. 新建任务，选中转站真实模型，不要选 Luna / gpt-reserve。

同一账号、同一套代理，IDE 不会锁到 gpt-reserve，中转可以正常用。

建议 CC Switch：转发 /responses 时按映射改写 gpt-reserve、gpt-5.6-luna。
建议 Codex 桌面端：Reserve 模式下不要藏自定义/路由模型，自定义 base_url 时不要强行改成 gpt-reserve。