Custom model metadata cannot be resolved although model_catalog_json is successfully loaded 已成功加载 model_catalog_json，但仍无法解析自定义模型元数据

Open 💬 8 comments Opened Jul 11, 2026 by xiandeyi

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of Codex CLI is running?

0.143.0

What subscription do you have?

Free

Which model were you using?

MiniMax-M2.7

What platform is your computer?

windows x64

What terminal emulator and version are you using (if applicable)?

powershell and cmd

Codex doctor report

What issue are you seeing?

Summary / 问题概述
English

When using a custom OpenAI-compatible provider (MiniMax) together with model_catalog_json, Codex successfully uses the custom model for inference, but still reports:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

Meanwhile:

/status correctly reports the custom model.
API requests succeed.
Tokens are consumed normally.
The custom model does not appear in /model.

This suggests that runtime model selection and metadata lookup are using different code paths.

中文

当使用 OpenAI Compatible Provider（MiniMax）并配置 model_catalog_json 后，Codex 已经能够正常调用自定义模型进行推理，但启动时仍然提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

与此同时：

/status 能正确显示当前模型；
API 请求完全正常；
MiniMax Token 正常消耗；
/model 中仍然没有该模型。

这表明模型调用成功，但模型元数据查询失败，两者似乎并未使用同一套模型目录。

Environment / 环境
English
OS: Windows 11
Codex CLI: v0.143.0
Provider: MiniMax (OpenAI Compatible)
API: Responses API
Configuration managed by CC Switch
Model: MiniMax-M2.7
中文
操作系统：Windows 11
Codex CLI：v0.143.0
Provider：MiniMax（OpenAI Compatible）
API：Responses API
使用 CC Switch 管理配置
模型：MiniMax-M2.7
Configuration / 配置
config.toml
model = "MiniMax-M2.7"
model_provider = "minimax"
model_context_window = 1000000

[model_providers.minimax]
name = "MiniMax"
base_url = "https://api.minimaxi.com/v1"
experimental_bearer_token = "***"
wire_api = "responses"
model_catalog_json = "C:/Users/<user>/.codex/model-catalogs/custom-catalog.json"
custom-catalog.json
{
"models": [
{
"slug": "MiniMax-M2.7",
"display_name": "MiniMax-M2.7",
"description": "MiniMax",
"default_reasoning_level": "high",
"supported_reasoning_levels": [
{
"effort": "none",
"description": "Think-Off"
},
{
"effort": "high",
"description": "Deep"
}
],
"shell_type": "shell_command",
"visibility": "list",
"supported_in_api": true,
"priority": 0,
"supports_reasoning_summaries": true,
"default_reasoning_summary": "none",
"supports_parallel_tool_calls": true,
"experimental_supported_tools": [],
"input_modalities": [
"text",
"image"
]
}
]
}
Reproduction / 复现步骤
English
Configure a custom OpenAI-compatible provider.
Configure model_catalog_json.
Set
model="MiniMax-M2.7"
Start Codex.
中文
配置 OpenAI Compatible Provider。
配置 model_catalog_json。
将
model="MiniMax-M2.7"

作为默认模型。

启动 Codex。

Actual Behavior / 实际表现
English

Codex starts successfully.

/status shows:

Model: MiniMax-M2.7
Model provider: MiniMax

Requests are successfully sent to MiniMax.

Token usage increases normally.

However, every startup prints:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

In addition, /model only lists built-in OpenAI models.

中文

Codex 能正常启动。

/status 显示：

Model: MiniMax-M2.7
Model provider: MiniMax

说明 Codex 实际已经在使用 MiniMax。

API 请求成功。

MiniMax Token 正常消耗。

但每次启动都会提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

另外，/model 中仅显示 OpenAI 官方模型，自定义模型不会出现。

Expected Behavior / 期望行为
English

Since:

the custom catalog is configured,
the model exists in the catalog,
/status already recognizes the custom model,
API requests succeed,

Codex should:

resolve metadata from model_catalog_json;
stop printing the warning;
include the custom model in /model.
中文

既然：

已配置 model_catalog_json；
自定义模型已存在于 Catalog；
/status 已能识别该模型；
API 请求已经正常；

那么希望 Codex：

能从 model_catalog_json 正确读取模型元数据；
不再输出 fallback metadata Warning；
/model 能显示该模型。
Investigation / 已完成排查
English

The following have been verified:

The provider configuration is correct.
The model is actually used for inference.
Tokens are billed by MiniMax.
model_catalog_json is configured correctly.
The custom catalog follows the documented schema.

Therefore this does not appear to be a provider configuration issue.

It appears that metadata lookup (get_model_info()) does not use the same runtime catalog that is used for model execution.

中文

已确认：

Provider 配置正确；
实际推理由 MiniMax 完成；
MiniMax Token 正常计费；
model_catalog_json 已配置；
自定义 Catalog 格式符合文档要求。

因此，这看起来不像 Provider 配置问题。

更像是：

模型执行路径已经能够识别自定义模型，而模型元数据查询 (get_model_info()) 并未使用同一份 Runtime Catalog。

Related Issues / 相关 Issue

This issue appears related to, but different from:

#24659
#19694

The difference is that the custom model is already active and being used successfully, but metadata lookup still falls back to built-in metadata.

This may indicate a regression in Codex CLI v0.143.0.

What steps can reproduce the bug?

Summary / 问题概述
English

When using a custom OpenAI-compatible provider (MiniMax) together with model_catalog_json, Codex successfully uses the custom model for inference, but still reports:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

Meanwhile:

/status correctly reports the custom model.
API requests succeed.
Tokens are consumed normally.
The custom model does not appear in /model.

This suggests that runtime model selection and metadata lookup are using different code paths.

中文

当使用 OpenAI Compatible Provider（MiniMax）并配置 model_catalog_json 后，Codex 已经能够正常调用自定义模型进行推理，但启动时仍然提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

与此同时：

/status 能正确显示当前模型；
API 请求完全正常；
MiniMax Token 正常消耗；
/model 中仍然没有该模型。

这表明模型调用成功，但模型元数据查询失败，两者似乎并未使用同一套模型目录。

作为默认模型。

启动 Codex。

Actual Behavior / 实际表现
English

Codex starts successfully.

/status shows:

Model: MiniMax-M2.7
Model provider: MiniMax

Requests are successfully sent to MiniMax.

Token usage increases normally.

However, every startup prints:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

In addition, /model only lists built-in OpenAI models.

中文

Codex 能正常启动。

/status 显示：

Model: MiniMax-M2.7
Model provider: MiniMax

说明 Codex 实际已经在使用 MiniMax。

API 请求成功。

MiniMax Token 正常消耗。

但每次启动都会提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

另外，/model 中仅显示 OpenAI 官方模型，自定义模型不会出现。

Expected Behavior / 期望行为
English

Since:

the custom catalog is configured,
the model exists in the catalog,
/status already recognizes the custom model,
API requests succeed,

Codex should:

resolve metadata from model_catalog_json;
stop printing the warning;
include the custom model in /model.
中文

既然：

已配置 model_catalog_json；
自定义模型已存在于 Catalog；
/status 已能识别该模型；
API 请求已经正常；

那么希望 Codex：

能从 model_catalog_json 正确读取模型元数据；
不再输出 fallback metadata Warning；
/model 能显示该模型。
Investigation / 已完成排查
English

The following have been verified:

Therefore this does not appear to be a provider configuration issue.

It appears that metadata lookup (get_model_info()) does not use the same runtime catalog that is used for model execution.

中文

已确认：

Provider 配置正确；
实际推理由 MiniMax 完成；
MiniMax Token 正常计费；
model_catalog_json 已配置；
自定义 Catalog 格式符合文档要求。

因此，这看起来不像 Provider 配置问题。

更像是：

模型执行路径已经能够识别自定义模型，而模型元数据查询 (get_model_info()) 并未使用同一份 Runtime Catalog。

Related Issues / 相关 Issue

This issue appears related to, but different from:

#24659
#19694

The difference is that the custom model is already active and being used successfully, but metadata lookup still falls back to built-in metadata.

This may indicate a regression in Codex CLI v0.143.0.

What is the expected behavior?

Summary / 问题概述
English

When using a custom OpenAI-compatible provider (MiniMax) together with model_catalog_json, Codex successfully uses the custom model for inference, but still reports:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

Meanwhile:

/status correctly reports the custom model.
API requests succeed.
Tokens are consumed normally.
The custom model does not appear in /model.

This suggests that runtime model selection and metadata lookup are using different code paths.

中文

当使用 OpenAI Compatible Provider（MiniMax）并配置 model_catalog_json 后，Codex 已经能够正常调用自定义模型进行推理，但启动时仍然提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.

与此同时：

/status 能正确显示当前模型；
API 请求完全正常；
MiniMax Token 正常消耗；
/model 中仍然没有该模型。

这表明模型调用成功，但模型元数据查询失败，两者似乎并未使用同一套模型目录。

作为默认模型。

启动 Codex。

Actual Behavior / 实际表现
English

Codex starts successfully.

/status shows:

Model: MiniMax-M2.7
Model provider: MiniMax

Requests are successfully sent to MiniMax.

Token usage increases normally.

However, every startup prints:

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

In addition, /model only lists built-in OpenAI models.

中文

Codex 能正常启动。

/status 显示：

Model: MiniMax-M2.7
Model provider: MiniMax

说明 Codex 实际已经在使用 MiniMax。

API 请求成功。

MiniMax Token 正常消耗。

但每次启动都会提示：

Model metadata for MiniMax-M2.7 not found.
Defaulting to fallback metadata.

另外，/model 中仅显示 OpenAI 官方模型，自定义模型不会出现。

Expected Behavior / 期望行为
English

Since:

the custom catalog is configured,
the model exists in the catalog,
/status already recognizes the custom model,
API requests succeed,

Codex should:

resolve metadata from model_catalog_json;
stop printing the warning;
include the custom model in /model.
中文

既然：

已配置 model_catalog_json；
自定义模型已存在于 Catalog；
/status 已能识别该模型；
API 请求已经正常；

那么希望 Codex：

能从 model_catalog_json 正确读取模型元数据；
不再输出 fallback metadata Warning；
/model 能显示该模型。
Investigation / 已完成排查
English

The following have been verified:

Therefore this does not appear to be a provider configuration issue.

It appears that metadata lookup (get_model_info()) does not use the same runtime catalog that is used for model execution.

中文

已确认：

Provider 配置正确；
实际推理由 MiniMax 完成；
MiniMax Token 正常计费；
model_catalog_json 已配置；
自定义 Catalog 格式符合文档要求。

因此，这看起来不像 Provider 配置问题。

更像是：

模型执行路径已经能够识别自定义模型，而模型元数据查询 (get_model_info()) 并未使用同一份 Runtime Catalog。

Related Issues / 相关 Issue

This issue appears related to, but different from:

#24659
#19694

The difference is that the custom model is already active and being used successfully, but metadata lookup still falls back to built-in metadata.

This may indicate a regression in Codex CLI v0.143.0.

Additional information

_No response_

View original on GitHub ↗

8 Comments

github-actions[bot] contributor · 1 month ago

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

#32119

Powered by Codex Action

zopiya · 1 month ago

我也遇见了，如下配置即可，需要写在顶层，可以尝试一下。配置后 codex 可以正常识别，没有报错。

─────┬────────────────────────────────────
     │ File: .codex/config.toml
─────┼────────────────────────────────────
   1 │ model = "MiniMax-M3"
   2 │ model_provider = "minimax"
   3 │ model_context_window = 1000000
   4 │ model_reasoning_effort = "high"
   5 │ approvals_reviewer = "user"
   6 │ model_catalog_json = "~/.codex/model-catalogs/custom-catalog.json"。     <------ 
   7 │
   8 │ [sandbox_workspace_write]
   9 │ network_access = true
  10 │ writable_roots = ["/Users/*****/******"]
  11 │ exclude_slash_tmp = true
  12 │ exclude_tmpdir_env_var = true
  13 │
  14 │ [model_providers.minimax]
  15 │ name = "MiniMax"
  16 │ base_url = "https://api.minimaxi.com/v1"
  17 │ experimental_bearer_token = "************"
  18 │ wire_api = "responses"

meesudzu · 1 month ago

got this too with same MiniMax provider

zopiya · 1 month ago

@meesudzu model_catalog_json should be a top-level configuration option rather than being nested under [model_providers.<provider>], since it applies globally instead of to a specific provider. 😶‍🌫️

meesudzu · 1 month ago

@meesudzu model_catalog_json should be a top-level configuration option rather than being nested under [model_providers.], since it applies globally instead of to a specific provider. 😶‍🌫️

Yes, for sure, the Codex CLI is applied, but the Codex app isn't.

Huynh-Phat-Gemera · 1 month ago

> @meesudzu model_catalog_json should be a top-level configuration option rather than being nested under [model_providers.], since it applies globally instead of to a specific provider. 😶‍🌫️ Yes, for sure, the Codex CLI is applied, but the Codex app isn't. <img alt="Image" width="607" height="200" src="https://private-user-images.githubusercontent.com/17334624/625433892-2b328df3-44a2-4723-b6e1-e2898b290d86.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3ODQ4MDA1NTksIm5iZiI6MTc4NDgwMDI1OSwicGF0aCI6Ii8xNzMzNDYyNC82MjU0MzM4OTItMmIzMjhkZjMtNDRhMi00NzIzLWI2ZTEtZTI4OThiMjkwZDg2LnBuZz9YLUFtei1BbGdvcml0aG09QVdTNC1ITUFDLVNIQTI1NiZYLUFtei1DcmVkZW50aWFsPUFLSUFWQ09EWUxTQTUzUFFLNFpBJTJGMjAyNjA3MjMlMkZ1cy1lYXN0LTElMkZzMyUyRmF3czRfcmVxdWVzdCZYLUFtei1EYXRlPTIwMjYwNzIzVDA5NTA1OVomWC1BbXotRXhwaXJlcz0zMDAmWC1BbXotU2lnbmF0dXJlPTQ0YWI5NjQ5ZjI1ZmZlNTJiMGQ1Y2QzMDAwODBkNWRmYTI3OTRmNTVjMWMwNTc4YjYxOWRjN2FlZWY5N2MyZjYmWC1BbXotU2lnbmVkSGVhZGVycz1ob3N0JnJlc3BvbnNlLWNvbnRlbnQtdHlwZT1pbWFnZSUyRnBuZyJ9.YjFu5MBXnjzyeUV99NetOw14XZmDAnNyG6wQDyRJYIQ"> <img alt="Image" width="401" height="92" src="https://private-user-images.githubusercontent.com/17334624/625434051-3eed9717-ee80-4e2b-b9a0-9ae176047775.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.p0FeIOVAvs3kiAaKVtVoy3SdBohNxO8z2Yj91jqvWZw">

https://github.com/openai/codex/issues/19694

The patch works on the desktop Codex, but not in VS Code.

zopiya · 1 month ago

> > @meesudzu model_catalog_json should be a top-level configuration option rather than being nested under [model_providers.], since it applies globally instead of to a specific provider. 😶‍🌫️ > > > Yes, for sure, the Codex CLI is applied, but the Codex app isn't. > <img alt="Image" width="607" height="200" src="https://private-user-images.githubusercontent.com/17334624/625433892-2b328df3-44a2-4723-b6e1-e2898b290d86.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.YjFu5MBXnjzyeUV99NetOw14XZmDAnNyG6wQDyRJYIQ"> <img alt="Image" width="401" height="92" src="https://private-user-images.githubusercontent.com/17334624/625434051-3eed9717-ee80-4e2b-b9a0-9ae176047775.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.p0FeIOVAvs3kiAaKVtVoy3SdBohNxO8z2Yj91jqvWZw"> #19694 The patch works on the desktop Codex, but not in VS Code.

I don't use the Codex GUI.

Maybe that's one of the pain points of vibe coding 😂

interconnectedMe · 13 days ago

I have a similar issue running V4 Flash subagent in Codex with gpt model acting as coordinator.
Works fine, but V4 Flash needlessly compacts using Codex gui's 258k context limit.
Tested max_context_window, model_auto_compact_token_limit, top-level model_context_window, and VS Code restart did not fix the spawned subagent's context up to 1M.