Batch API always returns `server_is_overloaded` and repeatedly processes same request

API Bugs
1

On 5/4/26, batch request hung on the last request and expired after 24 hours, returning the following error for the final request that failed:

{
  "id": "batch_req_69fa15d7ff648190afacfaf55a1755e6",
  "custom_id": "HIM610",
  "response": {
    "status_code": 503,
    "request_id": "39ac7fa7-deec-4b3f-a5f3-98bdc267fadd",
    "body": {
      "error": {
        "message": "Our servers are currently overloaded. Please try again later.",
        "type": "service_unavailable_error",
        "param": null,
        "code": "server_is_overloaded"
      }
    }
  },
  "error": null
}

The next day 5/5/26, testing with just one request in the batch payload, the request completed in 18 hours but still returning error:

{
  "id": "batch_req_69fb3c7a14e481909dfafafd63cf1f51",
  "custom_id": "HIM610",
  "response": {
    "status_code": 503,
    "request_id": "62df79e8-aa32-440a-916a-75cd09c549c0",
    "body": {
      "error": {
        "message": "Our servers are currently overloaded. Please try again later.",
        "type": "service_unavailable_error",
        "param": null,
        "code": "server_is_overloaded"
      }
    }
  },
  "error": null
}

Today, attempting again, the batch is still stuck in progressing, and I’m expecting it to complete or expire with the same server overload error.

The biggest issue besides the error is that OpenAI API is mysteriously processing the request repeatedly and consuming our budget. For example, the single payload with the lone request from yesterday 5/5/26 that completed but errored out was counted as 2 requests. It’s gotten worse today 5/6/26. The same exact payload with the lone request is, at the time of writing, up to 14 request count and the API is consuming from our budget repeatedly.

@OpenAI_Support @edwinarbus Please look into this and add credits back to our account for this issue. If unresolved, it’s going to just keep processing the same request in the batch payload while erroring out.

2

Hi,

This sounds related to another Batch API thread from yesterday, where the docs were linked and the 24h completion window / expired state behavior was discussed:

But since you’re also seeing the same single request counted/processed repeatedly and consuming budget, I’d definitely recommend you to open a support ticket for this if you haven’t already.

https://help.openai.com/en

I’d include the batch IDs, request IDs, timestamps, screenshots of the repeated request count/usage and the error payload showing server_is_overloaded.

Having a support ticket/case ID also makes it easier for OpenAI Support on the forum to follow up or connect the forum report to the account-specific issue.

Side note: @edwinarbus hasn’t been here since October.