feat(provider): add google-vertex-openai for Vertex AI partner models

[bluet]

Summary

Adds custom loader for google-vertex-openai provider to enable Vertex AI partner models (GLM, Llama, Qwen, DeepSeek) via the OpenAI-compatible Chat Completions API endpoint.

Problem

The @ai-sdk/google-vertex SDK only supports Google's Gemini models because it:

• Hardcodes publishers/google in URL construction
• Uses v1beta1 API version
• Expects Gemini-specific API format

Partner models like GLM-4.7, Llama 4, Qwen3, and DeepSeek use a different endpoint:

https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/endpoints/openapi/chat/completions

Solution

Created custom loader for google-vertex-openai provider that:

• Uses @ai-sdk/openai-compatible (already bundled in OpenCode)
• Injects Google OAuth tokens via custom fetch wrapper
• Uses google-auth-library for auto-refreshing tokens (transitive dep of @ai-sdk/google-vertex)
• Supports both global and regional endpoints
• Auto-loads when GOOGLE_CLOUD_PROJECT env var is set and credentials are available

Technical Details

Auth Method: Google Application Default Credentials (ADC)

• Uses GoogleAuth from google-auth-library
• Scope: https://www.googleapis.com/auth/cloud-platform
• Tokens auto-refresh 5 minutes before expiry

Endpoint Construction:

• Global: https://aiplatform.googleapis.com/v1/projects/{project}/locations/global/endpoints/openapi
• Regional: https://{location}-aiplatform.googleapis.com/v1/projects/{project}/locations/{location}/endpoints/openapi

Implementation Pattern:
Follows the same pattern as google-vertex-anthropic loader - custom SDK with OAuth injection.

Models Supported

Models are defined in companion PR: anomalyco/models.dev#716

• GLM-4.7 (zai-org)
• Llama 4 Maverick 17B-128E (meta)
• Llama 3.3 70B (meta)
• Qwen3 235B Instruct (qwen)
• DeepSeek V3.1 (deepseek-ai)

Testing

Requires:

• GOOGLE_CLOUD_PROJECT environment variable
• Google Application Default Credentials: gcloud auth application-default login
• Models enabled in Google Cloud Console (Model Garden)

Test command:

export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_LOCATION=global  # or regional like us-central1
opencode
# Select: google-vertex-openai/glm-4.7-maas

Related PRs

• feat: add google-vertex-openai provider for Vertex AI partner models models.dev#716 - Model definitions for google-vertex-openai provider

References

• Google Cloud: Call MaaS APIs for open models
• Google Cloud: OpenAI compatibility

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

POTENTIAL DUPLICATE FOUND:

PR #10303: feat: add google-vertex-openapi provider with ADC authentication
#10303

Why it's related:

• Both PRs are adding Google Vertex AI provider support with Application Default Credentials (ADC) authentication
• Both use OpenAI-compatible endpoints for partner models
• Both implement custom loaders with OAuth token injection
• The naming is nearly identical (google-vertex-openai vs google-vertex-openapi)

Recommendation: Check PR #10303 to see if it's addressing the same feature or if there's overlap in implementation. They may need to be consolidated or one may supersede the other.

[bluet]

Closing in favor of #10303, which takes a better approach (extending the existing google-vertex provider with a protocol: "openapi" field rather than creating a separate provider).

PR #10303 is more mature with comprehensive tests and aligns with the consolidation direction that @rekram1-node mentioned for models.dev. Happy to help review or test #10303 if needed.