The Batch API Decided to Make Hundreds of Requests That I Didn't Request. What's Up?

Well, I guess I’m going to give play by play updates on working with OpenAI Support here.

Three days ago I got this after expressing confusion over what HAR they were wanting and over their asking me to reproduce the batch request error:

Hello,​

Thank you for reaching back to OpenAI Support.

We can see the concern regarding being asked to reproduce the issue and the impact this may have on your usage costs.

To clarify, the intention is not to require repeated batch requests, but to establish clear confirmation that the issue occurred beyond expected system behavior. The previous experience may have been influenced by factors such as a batch expiring before completion, partial processing within the SLA window, or an interruption during output generation where no file was successfully returned despite internal processing activity.

In some cases, system logs may reflect internal states such as “no output” when the final result was not successfully generated or delivered, even if part of the request was processed. As outlined in the Batch API behavior, charges are applied to completed work within the batch, even if the full output is not returned due to expiration or failure to finalize.

The goal is to validate whether this instance reflects expected system handling or if it indicates behavior outside of normal conditions, so it can be properly reviewed.

Best,

Gesterly

OpenAI Support

Today I responded:

Hi, I am still confused at what you are wanting from me. I understand your intentions and the goal, but I am not able to replicate this.

As for this, “As outlined in the Batch API behavior, charges are applied to completed work within the batch, even if the full output is not returned due to expiration or failure to finalize,” what is the “Batch API behavior” you are referring to? Could you give a link to it?

Then I get an AI response and a new case number…

Case Number: 08359823

Hi, AI-assisted support here.

The “Batch API behavior” refers to the Batch API FAQ, specifically the sections on expired or canceled batches: if a batch expires or is canceled, completed work is returned and developers are charged for completed work.

https://help.openai.com/en/articles/9197833-batch-api-faq

The API reference for Batch is here:

https://platform.openai.com/docs/api-reference/batch

---

This response was generated with AI support which can make mistakes.

Why did they give me a new case number? Nothing was resolved previously.

Also, I am confused by the AI response, so I sent,

Ok, then I need all my money back. The Batch API FAQ says,

What happens if the API doesn’t complete my request within the promised time?

We aim to get all requests within 24 hours. If a batch expires (i.e. it could not be completed within the SLA time window), then remaining work is cancelled and any already completed work is returned. Developers will be charged for any completed work.

Which says completed work is returned, so if no work is returned, then there is no completed work. Thanks for the clarification. When should I expect the charges to be taken off of my account?

and got,

Case Number: 08359823

I can’t confirm a credit or removal of API charges without reviewing the specific batch and billing records.

Please send the batch ID(s), org/project ID, date/time in UTC, and the amount you’re disputing, and I’ll review what can be adjusted.

So, the bot cannot see my old case info that I was replying to. It’s escalated me to yet another “support specialist.” I wonder if I’ll every get this charge removed from my account.

@OpenAI_Support, is there any hope for me?