[BUG] API provider setting / GCP vertex AI

[Ats-F]

Problem (one or two sentences)

When using the GCP Vertex AI provider to access third-party MaaS models (e.g., gpt-oss-120b-maas, qwen, kimi-k2), the extension incorrectly prepends the publisher path as publishers/google/models/ instead of the correct publisher path (e.g., publishers/openai/models/). This results in a 404 NOT_FOUND error.

Context (who is affected and when)

Trying to use third-party or Model-as-a-Service (MaaS) models through the Google Cloud Vertex AI API provider in Roo Code.

Reproduction steps

1. Open Roo Code settings and select the "GCP Vertex AI" API provider.
2. Provide the necessary GCP credentials and project details.
3. Enter a third-party model name in the model field, such as gpt-oss-120b-maas (or qwen, kimi-k2).
4. Send any prompt in the chat.
5. Observe the 404 Not Found error indicating that the publisher model publishers/google/models/... was not found.

Expected result

The extension should route the request to the correct publisher path for the specified third-party model (for example, publishers/openai/models/gpt-oss-120b-maas for OpenAI-compatible models on Vertex AI), rather than forcing the google publisher prefix.

Actual result

The request fails because the publisher prefix seems to be hardcoded or defaulted to google. The generated path becomes publishers/google/models/, which leads to an immediate 404 error from the Google Cloud API.

Variations tried (optional)

No response

App Version

3.51.1

API Provider (optional)

Google Vertex AI

Model Used (optional)

gpt-oss-120b-maas (Also confirmed with qwen and kimi-k2)

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

{
  "error": {
    "timestamp": "2026-04-08T02:29:17.556Z",
    "version": "3.51.1",
    "provider": "vertex",
    "model": "gpt-oss-120b-maas",
    "details": "Gemini 生成コンテキスト ストリーム エラー: {\"error\":{\"message\":\"{\\n  \\\"error\\\": {\\n    \\\"code\\\": 404,\\n    \\\"message\\\": \\\"Publisher Model `projects/unique-moon-489106-r4/locations/us-central1/publishers/google/models/gpt-oss-120b-maas` was not found or your project does not have access to it. Please ensure you are using a valid model version. For more information, see: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions\\\",\\n    \\\"status\\\": \\\"NOT_FOUND\\\"\\n  }\\n}\\n\",\"code\":404,\"status\":\"Not Found\"}}"
  }
}

[roomote-v0]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause: When a user enters a third-party MaaS model name (like gpt-oss-120b-maas) in the GCP Vertex AI provider, two things go wrong:

1. VertexHandler.getModel() in src/api/providers/vertex.ts does not support arbitrary/custom model IDs -- if the model is not in the known vertexModels map, it falls back to the default model ID, ignoring the user's input.
2. The @google/genai SDK automatically constructs Vertex AI model paths as publishers/google/models/{model}, but third-party MaaS models require different publisher paths (e.g., publishers/openai/models/gpt-oss-120b-maas).

Proposed Solution:

1. Modify VertexHandler.getModel() (src/api/providers/vertex.ts) to support custom/unknown model IDs. When a model ID is not found in the known vertexModels map, instead of falling back to the default, pass through the user-provided model ID with a sensible default ModelInfo.

2. Support publisher-qualified model paths: Allow users to specify a full publisher-qualified model path (e.g., publishers/openai/models/gpt-oss-120b-maas) as the model ID. When the model string contains a /, pass it through as-is to the SDK. This is consistent with how Google's own documentation suggests using MaaS models, and the @google/genai SDK should support full resource names.

3. Update tests in src/api/providers/__tests__/vertex.spec.ts to cover these new code paths.

This approach gives users maximum flexibility to use any Vertex AI MaaS model by specifying the full publisher-qualified model path, without requiring new configuration fields.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[Ats-F]

@roomote I approve this plan. Support for publisher-qualified model paths is exactly what's needed to use third-party MaaS models on Vertex AI. Please proceed with the implementation. Thank you!

[roomote-v0]

Thanks for the approval @Ats-F! I've seen your comment and will proceed with the implementation now. I'll open a draft PR to share progress shortly.

[roomote-v0]

Opened a draft PR with my attempt at addressing this: #12078

The fix modifies VertexHandler.getModel() to pass through custom/unknown model IDs (including publisher-qualified paths like publishers/openai/models/gpt-oss-120b-maas) instead of silently falling back to the default model. Feedback welcome!