Hello,
I experienced something odd while using the Batch API for embeddings.
What happened:
I ran multiple batches of 2000 embedding requests.
When I monitor the API platform console, I see:
2000 completed, 0 failed of 2000 total requests
…but the generated error file actually contained several entries like this:
{
  "response": {
    "status_code": 0,
    "request_id": "",
    "body": {}
  },
  "error": null
}
For example, I only received 1,994 embeddings even though the dashboard showed everything as completed successfully. The 6 failed(?) ones don’t appear in the success file, only in the error file. The job summary still says “0 failed,” which is misleading. Also, not really sure what caused status code of 0.
It would be really helpful if:
- 
status_code: 0 requests were counted as failed in the batch summary.
 - 
The docs clarified what this status actually means (timeout? internal retry exhaustion?).
 - 
Optionally, the platform retried these automatically.
 
Anyone else seeing the same behavior? Would love insight from the OpenAI team on whether this is expected or being tracked.