[BUG]

[Makat0]

Problem (one or two sentences)

配置自定义模型无法生效

数据包中选择了一个其他的模型

Context (who is affected and when)

影响所有自定义模型名字使用

Reproduction steps

配置供应商，配置一个自定义的模型名字

Expected result

与自定义模型匹配

Actual result

与自定义模型不匹配

Variations tried (optional)

No response

App Version

v3.51.1

API Provider (optional)

None

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote-v0]

I'd like to help with this issue, but I need some clarification to ensure I implement the right solution. Could you please provide more details on the following:

• Which API provider are you using? (e.g., Anthropic, Gemini, OpenAI Native, Vertex, Bedrock, DeepSeek, Mistral, etc.)
• What custom model name did you configure? (the exact string you entered in the model field)
• What model name appeared in the actual API request? (the model ID shown in the network packet / second screenshot)

From my investigation, I can see a pattern in the codebase where several providers silently fall back to a default model when the user-configured model ID is not found in the predefined models map. For example in src/api/providers/anthropic.ts:

let id = modelId && modelId in anthropicModels ? (modelId as AnthropicModelId) : anthropicDefaultModelId

This affects providers including: Anthropic, Gemini, Vertex, Anthropic-Vertex, OpenAI Native, OpenAI Codex, Minimax, and all providers extending BaseOpenAiCompatibleProvider. If a custom model name is entered that does not match a known model key, it is ignored and the default model is used instead.

Knowing which specific provider you are using would help me scope the fix appropriately.

Please tag @roomote in your response with the answers, and I'll be happy to implement the fix once I have this information.

[Makat0]

@roomote 因为有一些三方的大模型网关会封装gpt、claude等模型，因此需要自定义BaseURL、modelId。
建议修复方案：当用户自定义modelId时，后端调用使用用户填写的，而不是使用default modelId，配置可以省略，让用户自己配置。这样可以方便用户自由的baseurl以及模型，方便使用任何的三方网关，只要遵循协议即可。

[roomote-v0]

Thanks for the clarification @Makat0! I understand the issue now - when a custom model ID is configured that doesn't match one of the predefined model IDs, the backend silently falls back to the default model instead of using the user-provided one. This affects users who want to use third-party gateways with custom base URLs and model IDs.

I'll work on a fix that ensures user-configured model IDs are used as-is when they don't match a known model, rather than falling back to the default. I'll open a PR to share progress.

[roomote-v0]

Opened PR #11943 with my attempt at addressing this. The fix ensures that user-configured custom model IDs are preserved and sent to the API as-is, rather than being silently replaced by the default model ID when they don't match a predefined model. Feedback welcome!