service_tier when using Vertex AI

[csanz91]

When using the @google/python-genai SDK with vertexai: true, the service_tier field in GenerateContentConfigOrDict has no effect. The SDK places service_tier in the JSON request body, but the Vertex AI API does not read flex/priority tier configuration from the request body. It uses HTTP headers (X-Vertex-AI-LLM-Request-Type and X-Vertex-AI-LLM-Shared-Request-Type).

The request with service_tier in the google-ai api works fine.

This exact same issue has been reported in: googleapis/java-genai#972 and googleapis/js-genai#1468

Error raised is:

    "code": 400,
    "message": "Invalid value at 'service_tier' (type.googleapis.com/google.cloud.aiplatform.v1beta1.ServiceTier), \"priority\"",
    "status": "INVALID_ARGUMENT",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.BadRequest",
        "fieldViolations": [
          {
            "field": "service_tier",
            "description": "Invalid value at 'service_tier' (type.googleapis.com/google.cloud.aiplatform.v1beta1.ServiceTier), \"priority\""
          }
        ]
      }
    ]
  }

[Venkaiahbabuneelam]

Hi @csanz91, Thank you for reporting this issue. To assist us in effectively investigating and reproducing it, please provide a minimal, runnable code example, including the complete error logs and SDK version. these things would be very helpful. Thanks!

[csanz91]

client = Client(
    vertexai=True,
    project="project-abc,
    location="global",
)

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="What's the weather like?",
    config=GenerateContentConfig(
        service_tier=ServiceTier.PRIORITY,
    )
)

[Venkaiahbabuneelam]

Hello @csanz91, Thank you for reaching out and providing such clear reproduction code. According to the documentation, the workaround is as follows: The SDK currently packages the service_tier configuration into the JSON request body. While the developer API accepts this, the enterprise API (Vertex AI) requires service tiers to be defined via HTTP headers.

"client = genai.Client(
vertexai=True, project='your_project_id', location='global',
http_options=HttpOptions(
api_version="v1",
headers={
"X-Vertex-AI-LLM-Request-Type": "shared",
"X-Vertex-AI-LLM-Shared-Request-Type": "priority"
},
)
)"

[csanz91]

I think this should still be treated as a bug rather than expected behavior. From a developer experience perspective, the main appeal of this unified SDK is that it abstracts away the underlying infrastructure differences between the developer API and Vertex AI.

Since GenerateContentConfig formally exposes service_tier as a high-level parameter, having to manually inject low-level HTTP headers the moment we flip vertexai=True feels like a leaky abstraction. The SDK's internal request builder should ideally intercept that parameter and map it to the correct headers for Vertex behind the scenes, keeping the cross-platform contract intact. (Especially since this same confusion is popping up in the Java and JS repos).