Spend tracking with Responses API + background enabled

[elabbarw]

Hello,

I need a method to track spend for long-running background jobs (azure/o3-deep-research) via the Responses API. When background mode is disabled, spend tracking behaves as expected. However, Deep Research jobs can run for 5–10+ minutes and spend data is harder to capture since the first response doesn't have any content.

Current workflow in my internal Open WebUI fork:

• POST /responses → returns resp_xxx job ID.
• Periodically call (backoff/retry,etc..) GET /responses/{id} until job status indicates completion.
• Capture the output payload.
• DELETE /responses/{id}` to remove the job.

Feature request:
Enable logging of GET requests with spend metrics when the following conditions are met:

• The request is to the Responses API with a job ID (resp_*)
• The job is complete and contains content

Maybe there's a better way - not sure... Alternatively, maybe LiteLLM can track the job until completion?

Goal:
To associate final spend values with each background job and have them recorded after the polling cycle.

Thanks,
Wanis

[elabbarw]

I noticed this #11640 but it doesn't seem to address the spend issue. As you can see the only entry with spend was the chat-based test