reasoning_tokens is calculated twice when streaming

[yf-yang]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

Check the code below. RUN usage outside's reasoning_tokens is twice of RUN usage's

If that async for event in response_stream is commented out , then the usage is correct.

Example Code

python
import asyncio

from dotenv import load_dotenv
from pydantic_ai import Agent, ModelRetry
from pydantic_ai.models.openai import OpenAIResponsesModel, OpenAIResponsesModelSettings

load_dotenv()

agent = Agent(
  model=OpenAIResponsesModel("gpt-5-nano"),
  model_settings=OpenAIResponsesModelSettings(
    openai_reasoning_effort="low",
    openai_service_tier="flex",
    max_tokens=8192,
    timeout=15,
  ),
)

retried = False


@agent.tool_plain
def foo():
  global retried
  if not retried:
    retried = True
    raise ModelRetry("Please retry this tool again")
  return "Success"


async def main():
  async with agent.iter("Calculate 100 * 200 / 3, then call tool foo") as run:
    async for node in run:
      if Agent.is_model_request_node(node):
        async with node.stream(run.ctx) as response_stream:
          async for event in response_stream:
            # print(event)
            pass
          print("RESPONSE usage", response_stream.get().usage)
          print("STREAM usage", response_stream.usage())
          print('RUN usage', run.usage())
        print("RUN usage outside", run.usage())

    print(run.result.all_messages())


if __name__ == "__main__":
  asyncio.run(main())

Python, Pydantic AI & LLM client version

1.0.6