Related GitHub Issue

Closes: #12074

Description

This PR attempts to address Issue #12074. Feedback and guidance are welcome.

Problem: When using the GCP Vertex AI provider with third-party MaaS models (e.g. gpt-oss-120b-maas, qwen, kimi-k2), the extension would silently fall back to the default model ID because the user-provided model was not in the known vertexModels map. This caused the SDK to construct a path like publishers/google/models/<default-model> instead of using the intended model.

Fix: Modified VertexHandler.getModel() to handle three cases:

1. Known model (in vertexModels map): use curated ModelInfo as before
2. Custom/unknown model: pass through the user-provided model ID as-is with sensible default ModelInfo. This supports both bare model names (gpt-oss-120b-maas) and fully-qualified publisher paths (publishers/openai/models/gpt-oss-120b-maas)
3. No model specified: fall back to the default (existing behavior)

Test Procedure

• Added 4 new test cases in vertex.spec.ts:
  • Custom model IDs pass through instead of falling back to default
  • Publisher-qualified paths (e.g. publishers/openai/models/...) pass through as-is
  • Tool preferences (edit/apply_diff) still apply to custom models
  • No model specified still falls back to default
• All 11 tests pass: cd src && npx vitest run api/providers/__tests__/vertex.spec.ts

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue.
• Scope: Changes are focused on the linked issue.
• Self-Review: Self-reviewed.
• Testing: New tests added covering custom model paths.
• Documentation Impact: No documentation updates required.
• Contribution Guidelines: Read and agreed.

Documentation Updates

• No documentation updates are required.

Additional Notes

The @google/genai SDK and Vertex AI API accept full resource names directly, so passing publishers/openai/models/gpt-oss-120b-maas as the model ID should route correctly to the third-party publisher endpoint.

Interactively review PR in Roo Code Cloud