›_ Codex Issues
Codex Issues / #30406

Codex App: GPT-5.5 fails with X-OpenAI-Internal-Codex-Responses-Lite while GPT-5.4 works

Open 💬 10 comments Opened Jun 28, 2026 by xingqi7979
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Summary

After updating Codex App on macOS, GPT-5.4 works normally, but GPT-5.5 fails immediately with the following error:

This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.

ChatGPT app works normally. The issue only happens in Codex App when selecting GPT-5.5.

Please confirm whether this is a Codex App routing issue or a GPT-5.5 rollout bug.

Environment

  • Product: Codex App desktop client
  • Platform: macOS
  • Codex App version: [26.623.42026]
  • macOS version: [26.4.1]
  • Model that fails: GPT-5.5
  • Model that works: GPT-5.4
  • ChatGPT app: Works normally

Steps to reproduce

  1. Open Codex App on macOS.
  2. Select model GPT-5.5.
  3. Send a simple task, for example:

Summarize this repository. Do not edit files.

  1. The task fails with:

This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.

  1. Switch the model to GPT-5.4.
  2. Send the same task again.
  3. The task works normally.

Expected behavior

GPT-5.5 should work in Codex App, or the app should not route GPT-5.5 through an unsupported Lite path.

Actual behavior

GPT-5.5 fails with the internal Lite channel error, while GPT-5.4 works normally.

Additional notes

This started after updating Codex App. The issue appears specific to Codex App and GPT-5.5. ChatGPT app works normally, and Codex App works when using GPT-5.4.

View original on GitHub ↗

10 Comments

github-actions[bot] contributor · 23 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #30403
  • #30238
  • #30388
  • #30224

Powered by Codex Action

Hespethorn · 22 days ago
  1. 关闭 Codex Desktop
  2. 打开文件资源管理器，导航到：

``
C:\Users\<你的用户名>\.codex\
``

  1. 找到并编辑 model_cache.json 文件
  2. 将 gpt-5.5部分"use_responses_lite": true 改为 "use_responses_lite": false
  3. 保存文件并重新启动 Codex Desktop
achieverForever · 22 days ago

Mac上修改model_cache.json 方法行不通，codex desktop启动时会自动覆盖。

horacecar · 21 days ago
Mac上修改model_cache.json 方法行不通，codex desktop启动时会自动覆盖。

你鎖住佢咪得，我已經解決左。

horacecar · 21 days ago
1. 关闭 Codex Desktop 2. 打开文件资源管理器，导航到： `` C:\Users\<你的用户名>\.codex\ ` 3. 找到并编辑 model_cache.json 文件 4. 将 gpt-5.5部分"use_responses_lite": true 改为 "use_responses_lite": false` 5. 保存文件并重新启动 Codex Desktop

Thanks Bro You Save My Day <33333333

Wellyfun · 21 days ago
> Mac上修改model_cache.json方法行不通，codex桌面启动时会自动覆盖。 你锁住渠咪得，我已经解决左边了。

怎么锁住渠道啊 我跟他一样情况 改完 再次打开又自动改回去了

Common1104 · 21 days ago

@Wellyfun 没用的，我改成false以后保存，把文件右键属性改为只读，确实是不会改为true了，但是还是报错，现在看下来只能把cc switch共享对话历史的功能关掉

gousiji0 · 21 days ago
  • 所属项目：全局/Codex Desktop 排障
  • 问题表现：Codex Desktop 与 codex exec 使用 gpt-5.5 时，报错 This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite
  • 原因定位：不是单纯模型名错误，而是 model_provider = "openai_http" 这条通道会走 Lite；在该通道下，gpt-5.5 会被服务端拒绝
  • 验证结果：gpt-5.5 + openai_http 失败；gpt-5.5 + openai 成功；gpt-5.4/gpt-5.4-mini + openai_http 也可成功
  • 正确做法：需要使用 gpt-5.5 时，将 C:\Users\Administrator\.codex\config.toml 中的 model_provider 切换为 openai
  • 后续避免规则：看到 Lite 不支持报错时，不要只检查 model，还要同时检查 model_provider；若必须保留 openai_http，则优先改用 gpt-5.4 或 gpt-5.4-mini
gxlmyacc · 21 days ago
所属项目：全局/Codex Desktop 排障 问题表现：Codex Desktop 与 codex exec 使用 gpt-5.5 时，报错 This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite 原因定位：不是单纯模型名错误，而是 model_provider = "openai_http" 这条通道会走 Lite；在该通道下，gpt-5.5 会被服务端拒绝 验证结果：gpt-5.5 + openai_http 失败；gpt-5.5 + openai 成功；gpt-5.4/gpt-5.4-mini + openai_http 也可成功 正确做法：需要使用 gpt-5.5 时，将 C:\Users\Administrator\.codex\config.toml 中的 model_provider 切换为 openai 后续避免规则：看到 Lite 不支持报错时，不要只检查 model，还要同时检查 model_provider；若必须保留 openai_http，则优先改用 gpt-5.4 或 gpt-5.4-mini

我试了下，将model_provider = "openai_https"改成model_provider = "openai"，重新打开codex，会报config.toml中的[model_providers.openai_https]下面的节点读取失败。并且原先的会话的都没了

harmonypiano · 17 days ago

found out that you can copy models_cache.json to model_catalog.json to make it persistent

then in your config.toml, add this (use absolute path):
model_catalog_json = "/absolute/path/to/model_catalog.json"