[Bug] ChatGPT account model list returns 400: /codex/models now requires client_version

[initH271]

What happened? / 问题描述

For the "OpenAI (ChatGPT Plus/Pro)" provider, the vendor-account model list request has failed with 400 since about 2026-09-24. PI-Desktop cannot get the account's model list, so any model that is only known from that list is rejected locally. For example, gpt-6.1-sol fails with MODEL_NOT_CONFIGURED: Vendor account does not offer model "gpt-6.1-sol" from session-launch.ts, and no request is sent upstream.

The logs only show vendor.account.model.list.failed … model list request failed (400), without the response body.

Root cause / 原因

apps/desktop/electron/main/vendor-live-models.ts (around L117–130) requests GET https://chatgpt.com/backend-api/codex/models with no query string. The endpoint now requires a client_version query parameter. The 400 body says:

('query','client_version') Field required

I tested the same account with 4 GET requests:

originator	query	status
pi	(none, same as PI-Desktop)	400: client_version Field required
pi	?client_version=0.159.2	200: 10 models, including gpt-6.1-sol
codex_cli_rs	?client_version=0.159.2	200: identical body
codex_cli_rs	(none)	400: same error

The originator header makes no difference. Only client_version matters.

Expected behavior / 预期行为

• Send client_version on the model-list request, as Codex CLI does.
• Include the upstream error body in the vendor.account.model.list.failed log entry, so problems like this are visible without an external repro.

Actual behavior / 实际行为

The model list is empty. Newer models such as gpt-6.1-sol cannot be used with the ChatGPT account, even though the same account serves them in Codex CLI.

App version / 应用版本

0.15.10 (macOS arm64). The request code is unchanged on main at 4547158.

[vastsa]

正在适配中，会引入pi新特性，官方更新冲突较多正在解决中