Batch API consistently expires without completion despite valid JSONL - Need investigation

Hi OpenAI team and community,

I’m experiencing a persistent issue where Batch API requests get stuck in in_progress status until they expire after 24 hours, despite using valid JSONL format and previously working configurations.

Issue Summary:

- Multiple batches submitted successfully but never complete

- Status progression: validating → in_progress → expired

- No errors returned, suggesting JSONL format is valid

- Issue started recently (previously worked fine)

Batch Submission Example:

curl -X POST “.…” \

-H “Authorization: Bearer $OPENAI_API_KEY” \

-H “Content-Type: application/json” \

-d '{

“input_file_id”: “file-92oSFSPtxQY3PW8sQ3AJjM”,

“endpoint”: “/v1/chat/completions”,

“completion_window”: “24h”

}’

Batch Status Response (Within 24h):

{

“id”: “batch_691c51b91e0c819081a4cd0055b2a4e7”,

“object”: “batch”,

“endpoint”: “/v1/chat/completions”,

“model”: “gpt-5-2025-08-07”,

“status”: “in_progress”,

“expires_at”: 1763550009,

“request_counts”: {

“total”: 1,

“completed”: 0,

“failed”: 0

},

“usage”: {

“input_tokens”: 0,

“output_tokens”: 0,

“total_tokens”: 0

}

}

Key Observations:

1. No error file generated - indicates JSONL is valid

2. Zero token usage - suggests processing never started

3. Multiple batches affected - not isolated to single request

4. Model: Using gpt-5-2025-08-07 (could this be related?)

Questions:

1. Is there a known issue with GPT-5 batch processing capacity?

2. Are there any current system-wide batch processing delays?

3. Any recommended debugging steps for stuck batches?

This is blocking our production insight generation system. Any guidance would be greatly appreciated!

Environment:

- Model: gpt-5-2025-08-07

- Endpoint: /v1/chat/completions

- Completion window: 24h

- Multiple batches affected over several days

- Problem started last night until now (before that, it worked fine)

Thanks in advance for any insights!