When using the Azure LLM key and employing "Runner.run_streamed", the token usage returned is 0.

[ch-royde]

Describe the bug
I replaced the key of OpenAI with Azure's key. I was able to correctly obtain the token usage information in the result returned by Runner.run(). However, when using Runner.run_streamed, the token usage was always returned as 0.

• openai: 1.84.0
• openai-agents: 0.0.17
• python: 3.10

client = AsyncAzureOpenAI(
    api_key=os.getenv("AZURE_OPENAI_API_KEY"),
    api_version=os.getenv("AZURE_OPENAI_API_VERSION"),
    azure_endpoint=os.getenv("AZURE_OPENAI_ENDPOINT"),
)

model = OpenAIChatCompletionsModel(
   model='gpt-4.1',
   openai_client=client,
)

agent=Agent(
  name="Agent",
  instructions=base_instructions,
  model=model,
  model_settings=ModelSettings(
     parallel_tool_calls=False, temperature=0.8
  ),
)

# Run streaming process
result = Runner.run_streamed(
    starting_agent=agent,
    input=enhanced_message,
    context=chat_context,
    max_turns=max_turns,
    run_config=run_config,
)

logger.info(f"Start stream events")
async for event in result.stream_events():
    if event.type == "raw_response_event":
        if isinstance(event.data, ResponseTextDeltaEvent):
            pass
    elif event.type == "run_item_stream_event":
        if event.item.type == "tool_call_item":
            logger.info(f"--- Tool Call: {event.item.raw_item.name} ---")
            logger.info(f"Tool Call Full Info: {event.item.raw_item}")
        elif event.item.type == "tool_call_output_item":
            logger.info("--- Tool Call Result ---")
            # Ensure output is string type
            output_str = str(event.item.output)
            # Only log first 1000 chars
            logger.info(output_str[:1000])

logger.info("**Stream successfully finished")
logger.info(f"**Final agent name: [{result.last_agent.name}]")
final_output = result.final_output
final_output_type = type(final_output)
logger.info(f"**Final output type: [{final_output_type}]")
logger.info(f"**Final output: [{final_output}]")

if final_output is None:
    logger.error("**Final output is None")
    raise NoneDataError(detail="Final output is None")

logger.info("**Token usage:")
total_input_tokens = 0
total_output_tokens = 0
total_tokens = 0
for index, r in enumerate(result.raw_responses):
    logger.info(
        f"{index}: input_tokens: [{r.usage.input_tokens}], output_tokens: [{r.usage.output_tokens}], total_tokens: [{r.usage.total_tokens}]"
    )
    total_input_tokens += r.usage.input_tokens
    total_output_tokens += r.usage.output_tokens
    total_tokens += r.usage.total_tokens
logger.info(
    f"**Total input tokens: [{total_input_tokens}], total output tokens: [{total_output_tokens}], total tokens: [{total_tokens}]"
)

final out logs: