Assistant API Cancellation Unexpectedly Continues Billing Usage Despite Run Being Cancelled

Below is some context on this issue regarding the cancellation of a thread and its billing outcome:

Understanding the Cancellation Process

From the provided JSON, the run shows a status of "cancelled", and the API returned a cancellation timestamp (e.g., "cancelled_at": 1699309538). This indicates that the cancellation request was accepted by the API. However, the fact that the request was still processing for several minutes in the UI and resulted in additional charges (50¢ of GPT-4-Turbo usage) suggests that different subsystems update on separate schedules.

Potential Factors at Play

• Delayed UI/Billing Sync:
  The cancellation of a thread may immediately report a "cancelled" status in the API response. However, the dashboard and billing systems might lag in reflecting the update. This can result in the appearance of continued processing and additional costs until all caches and billing records are updated.

• In-flight Processing:
  When a cancel is issued, there might be a short period where the cancellation is acknowledged but some operations (or charges associated with in-flight processing) are still finalized. This is common in asynchronous systems where multiple services (e.g., processing versus billing) reconcile their records over time.

Steps You Can Take

• Monitor Billing and Run State:
  Continue monitoring the thread’s state through the API. Once all related subsystems are updated, the UI should eventually reflect the cancellation accurately.

• Submit a Support Ticket:
  If the discrepancy persists or if you're unexpectedly billed for usage beyond the cancellation time, consider reaching out to OpenAI support with the thread ID and related timestamps. This will help verify if there was an internal delay or another issue in your cancellation workflow.

Example Reference JSON for a Cancelled Run

Below is your provided JSON (formatted for clarity) which confirms the cancellation:

JSONCopied
{
  "id": "run_gITrJ0bqR6mfNWR14Xq0AuyQ",
  "object": "thread.run",
  "created_at": 1699308938,
  "assistant_id": "asst_aoQqfTDFAKdxPCDciylDjnh9",
  "thread_id": "thread_Wpd3llJKfEbno4lNuZ4vDVd3",
  "status": "cancelled",
  "started_at": 1699308938,
  "expires_at": null,
  "cancelled_at": 1699309538,
  "failed_at": null,
  "completed_at": null,
  "last_error": null,
  "model": "gpt-4-1106-preview",
  "instructions": "<prompt>",
  "tools": [
    {
      "type": "retrieval"
    }
  ],
  "file_ids": [
    "file-PMI1PCpjx5iPy5YkE0EGUlvp",
    "file-XsnJaOGje94jn1mwoKbO52dV",
    "file-7wbACFrcvyrPRG7ZGrpcbEqP",
    "file-gKwDiZLOJDMPf5MoMkpsZMPL"
  ],
  "metadata": {}
}

Summary

Although the cancellation was confirmed in the API response, the delay in UI and billing update might have caused the observed charges and extended processing time. Reviewing all logs and possibly contacting support with the detailed thread and run IDs will help resolve whether this is a transient delay or something that needs further investigation.

Let me know if you need additional details or have further questions on managing these cancellations through the API.