// event · sdk
openai-python 3.36.0 adds Vertex AI Responses API support and custom voice consent
sdkvendor: openaipublished: 2h agoconfidence: 0.86id: 8cae4775-c7ef-42e6-8ac2-1def3927c30a
The official OpenAI Python SDK v3.36.0 adds Vertex AI as a backend for the Responses API, exposes 30-minute cache-write usage counts, surfaces compaction trigger IDs, introduces custom voice consent management APIs, and preps a native synchronous WebSocket option type. Also ships numerous WebSocket/Realtime stability fixes and preserves Vertex API roots across response lifecycles.
capability delta
An agent can now call the OpenAI Responses API through Vertex AI, track 30-minute cache-write token usage, read compaction trigger IDs, and manage custom-voice consent from Python.
curator note
Headline change is Vertex AI backend support for Responses API — meaningful for multi-cloud agent deployments. Voice consent management is also new surface area.
// install
pip install openai==3.36.0pypi · openai @ 3.36.0
// example
python
from openai import OpenAI
# Route the Responses API through Vertex AI
client = OpenAI(
base_url="https://aiplatform.googleapis.com/v1/projects/my-proj/locations/us-central1",
api_key="<vertex-access-token>",
)
resp = client.responses.create(
model="gpt-5",
input="Summarize the latest Vertex AI integration.",
)
print(resp.output_text)
# Inspect 30-minute cache-write usage and compaction trigger IDs
print(resp.usage.input_tokens_details)
print(getattr(resp, "compaction_trigger_id", None))
// links