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

A developer discovered a workaround for a bug in Codex Desktop where requests fail with the error 'This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.' The issue occurs after updating Codex Desktop on Windows 11, while ChatGPT web and desktop apps work normally. The temporary fix involves editing the local cache file 'model_cache.json' to set 'use_responses_lite' to false, though Codex automatically reverts this change on subsequent requests.

Language:English / 中文 Status:Temporary Workaround Not an Official Fix After updating Codex Desktop, every request immediately failed with the following error: This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite. The issue only occurred in Codex Desktop . The ChatGPT desktop/web app continued to work normally. - Windows 11 - Codex Desktop - GPT-5.5 - ChatGPT Plus - GPT-5.5 Medium ❌ - GPT-5.5 High ❌ - Every prompt failed immediately - ChatGPT worked normally - Restarting Codex did not help - Reinstalling Codex did not help After checking the following GitHub issues: - 30224 - 30406 I found that modifying the local cache file: %USERPROFILE%\.codex\model cache.json from "use responses lite": true to "use responses lite": false allowed Codex Desktop to work again. However, after sending a new request, Codex automatically rewrote model cache.json and restored "use responses lite": true This means that manually editing the file is not a permanent solution. Based on the observed behavior, model cache.json appears to be a runtime capability cache rather than a user configuration file. It seems that Codex regenerates this file whenever it refreshes model metadata from the server. Because of this, manual changes are overwritten automatically. This is an observation rather than an officially confirmed implementation. - Close Codex Desktop completely. - Open %USERPROFILE%\.codex\model cache.json - Change "use responses lite": true to "use responses lite": false - Save the file. If Codex keeps overwriting the file, you may temporarily mark it as Read-only . This is only a temporary workaround. Since the file is regenerated automatically, future updates may overwrite it again. Once OpenAI fixes the issue, this workaround should no longer be necessary. - GitHub Issue 30224 - GitHub Issue 30406 更新 Codex Desktop 后，无论发送什么 Prompt，都会立即报错： This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite. 该问题仅发生在 Codex Desktop 中。 ChatGPT（网页端/桌面端）可以正常使用。 - Windows 11 - Codex Desktop - GPT-5.5 - ChatGPT Plus - GPT-5.5 Medium ❌ - GPT-5.5 High ❌ - 所有请求立即失败 - ChatGPT 正常 - 重启 Codex 无效 - 重装 Codex 无效 查阅 GitHub Issue： - 30224 - 30406 发现修改本地文件： %USERPROFILE%\.codex\model cache.json 将 "use responses lite": true 修改为： "use responses lite": false 之后 Codex 可以恢复正常。 但是发送下一条消息后，Codex 又会自动把： "use responses lite": false 恢复成： "use responses lite": true 说明： model cache.json 不是用户配置文件（Configuration） ， 而是 Codex 自动维护的运行时缓存。 根据目前观察到的行为来看， model cache.json 更像是一个 模型能力缓存（Capability Cache） 。 Codex 会根据服务器返回的模型元数据自动重新生成这个文件，因此： - 手动修改能够立即生效； - 但后续会被自动覆盖。 目前这只是根据现象作出的分析，并非 OpenAI 官方确认的实现细节。 关闭 Codex。 打开： %USERPROFILE%\.codex\model cache.json 将： "use responses lite": true 改成： "use responses lite": false 保存即可。 如果 Codex 反复覆盖，可以暂时将该文件设置为 只读 。 该方案只是临时绕过（Workaround），并不是官方修复。 由于 model cache.json 会自动生成，因此未来版本可能重新覆盖，也可能官方修复后无需再进行此修改。 - GitHub Issue 30224 - GitHub Issue 30406