How to access cached tokens count from Usage Object

API Bugs
, ,
1

Hey Team,

I’m working with the OpenAI Agents SDK and noticed that the agent response includes a Usage object. This object shows properties that help me understand the model’s input and output token counts for the agent run. However, I don’t see any way to access cached tokens.

Here is the code snippet to run agent and access usage value

result = await Runner.run(
        starting_agent = demo_agent, 
        input = input
    )
    # Print the original final_output
    print("Final output:", result.final_output)
    print(result.raw_responses[0].usage)

and I can see the usage printed as below

usage=Usage(requests=1, input_tokens=935, output_tokens=15, total_tokens=950)

Is there a specific reason why cached tokens aren’t included in the Usage object? Do I need to enable any additional settings to have them reflected?

2

It works for me (openai-agents 0.2.9):
Usage(requests=1, input_tokens=22, input_tokens_details=InputTokensDetails(cached_tokens=0), output_tokens=15, output_tokens_details=OutputTokensDetails(reasoning_tokens=0), total_tokens=37)

Have you tried upgrading your packages?
pip install --upgrade openai openai-agents