Problem Summary:
1.
Duplicate Tasks: A single API request was replicated and executed multiple times on OpenAI’s servers.
2.
Request Timeout: Although the tasks were processed on the backend, the original API call never completed and was terminated by our timeout setting.
3.
Suspected Root Cause: We prepended a high-precision timestamp to the input content for tracking purposes. By comparing the OpenAI backend logs, we found that all duplicate tasks contained the exact same timestamp in their input. This serves as strong evidence that the duplication was not caused by our client-side logic (such as a retry mechanism) but rather by an internal failure within the OpenAI API service.
4 Likes
I’m seeing the same thing.
Model: o3-deepresearch.
The created entries show up a few minutes apart, so it looks like the same request ran multiple times. On top of that, outputs were empty and the calls timed out, but I still got charged for the duplicates—totaled around ~$1,000.
I’m requesting official support and a refund (credit).
1 Like
Hey @tgw13 and @founder6, are you still blocked / owed refunds here? Let me know if so, and I can look into your Support requests to make sure we resolve things for you!
1 Like
@OpenAI_Support
Thank you. I tried to contact support via chat, but the automated responses didn’t address my issue, so I had to stop partway through. As of now, my refund request (and related requests) has not been processed.
Could you let me know the appropriate channel and steps for submitting a support request?
@tgw13 Just sent you an email!
1 Like
I had the same issue with gpt-5-mini and still haven’t received a response back. I was getting the same issue as of earlier this week and have been trying to work around this issue by running 1 small batch at a time which defeats the purpose of batch processing.