`required` fields in `response_json_schema` are not enforced - API accepts responses missing required fields

[chrissykrissy]

Description

When using structured JSON output with response_json_schema, the required constraint in the schema is not enforced. The API accepts and returns responses that are missing fields marked as required without raising an error.

Environment

• SDK: google-genai
• Backend: Vertex AI
• Model: gemini-3-flash-preview
• Python version: 3.11+

Minimal Reproducible Example

from google import genai
from google.genai import types

client = genai.Client(vertexai=True, project="your-project", location="us-central1")

# Schema requires "prompt" field
schema = {
    "type": "object",
    "properties": {
        "prompts": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "prompt": {"type": "string"},
                    "content": {"type": "string"},
                    "confidence": {"type": "number"},
                },
                "required": ["prompt"],  # <-- "prompt" is REQUIRED
            },
        }
    },
    "required": ["prompts"],
}

# System prompt example shows "content" field (not "prompt")
user_prompt = '''
Generate a prompt in this JSON format:
{
  "prompts": [
    {
      "content": "Your generated text here",
      "confidence": 0.9
    }
  ]
}
'''

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=user_prompt,
    config=types.GenerateContentConfig(
        response_mime_type="application/json",
        response_schema=schema,
    )
)

print(response.text)
# Returns: {"prompts": [{"content": "...", "confidence": 0.9}]}
# Note: "prompt" field is MISSING but no error is raised!

Expected Behavior

One of the following:

1. The API should enforce the schema and ensure required fields are present
2. OR raise a validation error when the response doesn't include required fields
3. OR document that required is advisory only, not strictly enforced

Actual Behavior

The API silently accepts responses missing required fields.

Impact

In production, Gemini sometimes returns responses with only content while our schema requires prompt. Since no error is raised, invalid responses propagate through the system causing hard-to-diagnose bugs.

Additional Context

• Behavior is intermittent - sometimes responses include all fields, sometimes they don't
• The model appears to follow natural language instructions over schema constraints
• When schema says required: ["prompt"] but the prompt example shows "content", the model returns content only

[navneetkumaryadav207001]

I believe this could be addressed in one of two ways:

Provide an optional validator API

• Developers could call something like genai.validate_response(response, schema) to check compliance.

Add a strict (or validation_mode) parameter in GenerateContentConfig

• When strict=True, the backend validates the response and raises an error if required fields are missing.

I’d like to take ownership of implementing a solution once we agree on the approach. This would clarify the contract for developers and prevent silent propagation of invalid JSON, while still allowing flexible generation when desired.