Why are my batch jobs stuck at the "batch finalizing" stage?

Question

Why are my batch jobs stuck at the "batch finalizing" stage?

Pedro 40

I submitted a few very small test batch jobs using o4-mini and gpt5. However, they are currently stuck in the "batch finalizing' stage for over 48 hours. Given that "Batch in progress" is checked, I'm assuming that the model has finished processing the data. So then, why is it taking for the batch to return the results? Is there something I can do about this?

Pedro 40 Reputation points

2026-02-23T18:06:03.2033333+00:00

I eventually got this error:

Validation failed: An internal error has occurred. If any of the input for this batch was processed before the internal error occurred, these partial results will be available at the output location

Why did this happen? is there anything I could have done to prevent this?

Are we charged for this? It is quite the hassle to have to sort through all the partial completions to find what needs to be rerun....
Anshika Varshney 7,815 Reputation points Microsoft External Staff Moderator

2026-02-24T10:29:03.92+00:00
**Hi Pedro,
**Thanks for sharing the details, this one is unfortunately a known pattern with batch processing.

This error typically indicates a transient service‑side failure that occurred after the batch had already started executing. Some inputs were successfully processed, but the backend encountered an internal issue partway through execution and stopped the remaining work. The fact that retries sometimes succeed with the same input strongly suggests the failure is not caused by your data, but by temporary backend instability in the batch execution pipeline.

There isn’t anything you could have done to fully prevent it. These errors are not due to request formatting or configuration. The usual mitigations are operational rather than preventative:

Use smaller batch sizes so fewer items are affected when a failure occurs.

Implement retry logic for only the missing or failed inputs.

Design downstream handling assuming partial outputs are possible, as the service explicitly allows partial completion when failures occur.

You are generally charged only for the inputs that were actually processed, not for the items that never ran. The presence of partial outputs means those requests reached execution, so they count toward usage. For exact confirmation at the subscription level, checking usage metrics.

Totally agree on the hassle sorting partial completions is painful. Splitting batches and tracking completed IDs tends to reduce rework significantly when these transient failures happen.

1 answer

Your answer

Pedro 40 Reputation points

2026-02-23T18:06:03.2033333+00:00

I eventually got this error:

Validation failed: An internal error has occurred. If any of the input for this batch was processed before the internal error occurred, these partial results will be available at the output location

Why did this happen? is there anything I could have done to prevent this?

Are we charged for this? It is quite the hassle to have to sort through all the partial completions to find what needs to be rerun....
Anshika Varshney 7,815 Reputation points Microsoft External Staff Moderator

2026-02-24T10:29:03.92+00:00

**Hi Pedro,
**Thanks for sharing the details, this one is unfortunately a known pattern with batch processing.

This error typically indicates a transient service‑side failure that occurred after the batch had already started executing. Some inputs were successfully processed, but the backend encountered an internal issue partway through execution and stopped the remaining work. The fact that retries sometimes succeed with the same input strongly suggests the failure is not caused by your data, but by temporary backend instability in the batch execution pipeline.

There isn’t anything you could have done to fully prevent it. These errors are not due to request formatting or configuration. The usual mitigations are operational rather than preventative:

Use smaller batch sizes so fewer items are affected when a failure occurs.

Implement retry logic for only the missing or failed inputs.

Design downstream handling assuming partial outputs are possible, as the service explicitly allows partial completion when failures occur.

You are generally charged only for the inputs that were actually processed, not for the items that never ran. The presence of partial outputs means those requests reached execution, so they count toward usage. For exact confirmation at the subscription level, checking usage metrics.

Totally agree on the hassle sorting partial completions is painful. Splitting batches and tracking completed IDs tends to reduce rework significantly when these transient failures happen.

Answer 1

Hi Pedro,

Thanks for raising this. When batch jobs remain in “batch finalizing” for an extended period, it’s usually not related to the request size or client configuration. We’ve seen similar behavior during backend service delays where processing completes but result collation is slow or blocked.

A few things you can check in the meantime:

Review the batch metadata (for example, any error_file_id) to confirm whether the job finalized with warnings or errors.
Check the Azure Status page for any active or recent Azure OpenAI service issues affecting batch workloads.
If the job has been stuck for a long time, cancelling and resubmitting the batch can help once the service backlog clears.

If the issue persists across multiple submissions or regions, opening an Azure Support ticket with the batch IDs and timestamps would be the best next step so the team can investigate on the service side.

Thankyou!

Share via

Why are my batch jobs stuck at the "batch finalizing" stage?

1 answer

Your answer