OpenAI Agent SDK – Token usage and reasoning output

API
1

I’m currently using the OpenAI Agent SDK with an Azure-hosted reasoning model (o1). The setup is working well, and I’m successfully receiving responses from the Agent via chat completions.

In the response object, I can see the usage node which includes both output_tokens and reasoning_tokens.

I have a couple of questions for this:

  1. Is the output_tokens count inclusive of reasoning_tokens?
    I’m trying to determine the actual token usage cost, and understanding this breakdown will help.
  2. Is there any way to configure or enforce the model not to emit reasoning_tokens in the Agent’s response?

Any insights, documentation pointers, or best practices would be highly appreciated. Thanks in advance!

2

On the OpenAI API:

  • output_tokens is the total billing for AI generation
  • reasoning_tokens is the portion of that unseen by being internal

The usage details in the API object return, which contains reasoning_tokens, along with other categories like cached amount of input, can just be ignored if you don’t find the information useful to present.