Skip to content

Missing supports_reasoning_summaries field - Codex fails to start

Quick fix

Upgrade cc-switch to v3.18.0 or newer, which fixes required fields missing from the generated model catalogue and stopping Codex from starting.

Error output
切换供应商后生成的 cc-switch-model-catalog.json 中,DeepSeek 模型缺少 supports_reasoning_summaries 字段,导致 Codex 启动失败
English translation

In the cc-switch-model-catalog.json generated after switching provider, DeepSeek models lack the supports_reasoning_summaries field, so Codex fails to start

cc-switch fails to fill in the `supports_reasoning_summaries` field automatically for Chat Completions providers such as DeepSeek when generating `cc-switch-model-catalog.json`, so Codex (0.144.5+) fails to start while parsing the catalogue. In v3.18.0 the generation logic was updated to backfill the fields the parser requires from a built-in static template.

Worth noting: this issue cluster actually mixes several independent DeepSeek-related problems. Besides Codex failing to start: 1. cc-switch cannot intercept the Codex OpenAI quota query endpoint, producing a white screen (#3615, no fix yet); 2. adding a DeepSeek provider in the UI triggers a white screen (#3796, officially fixed in v3.16.4, though some users still reproduce it on v3.17.0); 3. taking over the Hermes config appends `custom_providers` repeatedly, producing a duplicate YAML key error (#4130, fixed in v3.16.3).

  1. Upgrade cc-switch to v3.18.0 or newer.

  2. Switch or save the DeepSeek provider config once more in cc-switch to trigger regeneration of `cc-switch-model-catalog.json`.

ToolCodex
Version0.144.5+
PlatformsWindowsmacOS
Adding a DeepSeek provider still shows a white screen after upgrading to v3.17.0. What now?
The form was reworked officially in v3.16.4. If a newer version still shows a white screen, open developer tools to see the console error and report back with reproduction steps and logs.
I have a DeepSeek API — why does Codex still say the quota is insufficient, or show a white screen?
At startup, Codex queries the OpenAI entitlements endpoint independently. cc-switch can currently only hijack the model inference endpoint, not the quota query, so the front end still blocks on an OpenAI account quota of 0 even with a third-party API configured.

This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.