IndexError in accumulate_event during Responses API streaming

Bug Report

Description

IndexError: list index out of range occurs inside the OpenAI SDK during streaming of Responses API events. The error originates in openai/lib/streaming/responses/_responses.py line 344 in accumulate_event, called from handle_event (line 248) and __stream__ (line 156).

This is not caused by application code — it happens internally within the SDK's stream accumulator.

Stack Trace

File "/app/.venv/lib/python3.13/site-packages/openai/lib/streaming/responses/_responses.py", line 151, in __aiter__
File "/app/.venv/lib/python3.13/site-packages/openai/lib/streaming/responses/_responses.py", line 156, in __stream__
File "/app/.venv/lib/python3.13/site-packages/openai/lib/streaming/responses/_responses.py", line 248, in handle_event
File "/app/.venv/lib/python3.13/site-packages/openai/lib/streaming/responses/_responses.py", line 344, in accumulate_event
    IndexError: list index out of range

Reproduction Context

• SDK version: openai==2.16.0
• Python version: 3.13
• API: Responses API with streaming (stream=True)
• Model: gpt-5-nano
• Features used: tools=[{"type": "web_search"}], structured output with text_format (Pydantic model)
• Streaming method: AsyncOpenAI().responses.stream() and resume via responses.stream() with starting_after

Code Pattern

async with client.responses.stream(...) as stream:
    async for event in stream:  # IndexError thrown here intermittently
        ...

The error occurs intermittently — most streams complete fine, but occasionally the accumulator hits an index that doesn't exist in its internal list.

Workaround

Wrapping the stream iteration in a try/except IndexError to catch and log the error, since the stream has typically delivered the important data by the time this occurs.

Notes

A similar class of bug exists in the Assistants API streaming path (_assistants.py in accumulate_event) — see #1736 and #1882. This issue is for the newer Responses API streaming path (responses/_responses.py).

Start by reading openai/lib/streaming/responses/_responses.py, especially accumulate_event at line 344 and its callers handle_event and stream. Reproduce the intermittent failure with AsyncOpenAI().responses.stream() using the Responses API context described, then verify that streaming and resuming with starting_after complete without an IndexError.