CitationMetadata model deviates from docs

[alexandersteinhauer]

GenerationContentResponse coming from the Batch API contains CitationMetadata as specified in the docs: https://ai.google.dev/api/generate-content#citationmetadata

class CitationMetadata:
    citationSources: list[CitationSource]

The library models it different though, leading to a mismatch:

class CitationMetadata:
    citations: list[Citation]

As CitationMetadata also inherits from _common.BaseModel all extra properties lead to an exception.
And because (at least) the Batch API returns the "correct" model, parsing it using the library is not possible.

Environment details

• Programming language: Python
• OS: Linux
• Language runtime version: Python 3.13
• Package version: 1.28.0

Steps to reproduce

1. Download jsonl output file from Batch API that has response.candidates.0.citationMetadata.citationSources populated.
2. Try to parse response with GenerateContentResponse.model_validate_json

[sharvil]

We got bit by this as well. We're not using the field, so we ended up with:

            response_obj = json.loads(response_json)
            # Server returns some stuff that doesn't validate with Pydantic so we have to filter
            # it out here.
            for c in response_obj['response']['candidates']:
              if 'citationMetadata' in c:
                del c['citationMetadata']  # Remove citation metadata if present

[janasangeetha]

Hi @alexandersteinhauer
Apologies for responding late. We truly appreciate you flagging this issue, we will file a bug internally.