Since today (2026-09-02) a share of our background-mode Responses API requests never leave the queued state. They are accepted with HTTP 200 and status: “queued”, but they never move to in_progress or completed, even after several hours. Polling GET /v1/responses/{id} keeps returning queued, with error: null, incomplete_details: null and usage: null. The dashboard logs show no output for them either.
Stuck requests (still queued at the time of writing, times UTC):
resp_0a1d93feea04e752006a983cf9603887d288a2d9756da6fa9d, created 15:12:58. Production request: text plus 4 image inputs, strict JSON schema, max_output_tokens 120000, reasoning effort medium. Queued for over 5 hours.
resp_09b00f8e128f9509006a987a9bda5487d284f7a7adbe0d5fdd, created 19:35:56. Same request shape. Our client gave up after 600 seconds; a retry with an identical payload was picked up and completed in about a minute.
resp_08bd7e48def8656d006a987df0cc8487d2abe345aed544698f, about 20:03. Minimal probe: “Reply with the single word OK” plus 4 image inputs, max_output_tokens 64, reasoning low.
resp_014dcebcd0788b8e006a987e4adb0087d28e12fcadf65e85c6, about 20:08. Minimal probe: text only, no images, no schema, max_output_tokens 64, reasoning low.
Requests with identical or heavier shapes that completed within 5 seconds, same project, same model, same time window:
resp_05aba7359f89f875006a987db6c84c87d2ac444600db3a090d (tiny text prompt)
resp_02a997a280d01694006a987dbe0ee887d2b35a48288265c314 (control on gpt-5.5)
In a final test between roughly 20:08 and 20:10 UTC we submitted eight identical minimal text-only background requests two seconds apart; two of the eight stayed queued, six completed within seconds. Synchronous requests (background omitted) to the same model complete in about one second every time, including with the same image inputs.
So the payload shape is not the cause: identical minimal requests are scheduled or not at random, and the unscheduled ones never recover. The status page shows no incident.
Would be nice if someone could confirm they have the same issues..
Having the same issues. We fall back to the Chat Completions API after 20mins in queued for our use case, and that has spiked quite a lot in the past couple days.
Main Model: gpt-5.4, Responses API in background mode
Fallback model: gpt-4o, Chat Completions API
Date
Total tasks
Responses API timeouts
Fallback attempts
Fallback successes
Timeout rate
08-15
311
0
0
0
0%
08-16
373
0
0
0
0%
08-17
1220
0
8
1
0%
08-18
1415
0
0
0
0%
08-19
1258
1
1
1
0.08%
08-20
1155
0
0
0
0%
08-21
1000
1
1
1
0.1%
08-22
332
2
2
2
0.6%
08-23
342
0
0
0
0%
08-24
1158
0
6
6
0%
08-25
991
1
40
7
0.1%
08-26
907
9
9
9
1.0%
08-27
1003
0
0
0
0%
08-28
970
0
0
0
0%
08-29
275
0
0
0
0%
08-30
369
0
0
0
0%
08-31
1197
2
3
3
0.2%
09-01
1055
86
86
86
8.2%
09-02 (partial, today)
848
123
121
121
14.5%
There is an issue marked “Resolved” yesterday in the Responses API, but I am still seeing the above today more and more frequently.
We are observing the same issue since approximately 06:00 UTC on 2 September.
Our requests use the Responses API with background: true.
Affected requests are accepted successfully, but remain in queued during repeated retrieve calls.
Each retrieve call returns HTTP 200, the response does not transition to in_progress or completed, and no provider error is returned.
We observed this first with gpt-5.4-2026-03-05 and also with gpt-5.6-terra using reasoning.effort: low.
We poll once every five seconds for up to 120 attempts, approximately ten minutes. The affected requests remain queued throughout that period.
The issue is still occurring at the time of posting, although the frequency appears to have decreased compared with the initial impact.
We’ve also been seeing the same issue with the Responses API, starting at 1st Sep 2026, 17:38 UTC and continuing to affect us at time of posting (3rd Sep 2026, 16:30 UTC).
Affected Responses simply stay in the “queued” status indefinitely, never completing or erroring:
Oh my god, finally seeing some others users encountering the same issue !
For us, this started 2 days ago (Sep 1 2026).
I hope they will solve this soon, is as it is currently blocking our workflows.
I have created a very similar topic a few hours ago.
Batch batch_6a9a35171f908190af12affae60e21fa (438 requests, gpt-5.6-luna, created 2026-09-04 03:03 UTC) has been hanging in in_progress at 434/438 completed for over 4 hours, with the last request stuck since ~09:40 UTC.
hi all, i also experienced this, there was only 1 item (dummy) in my batch and it also took full 24 hours and then there was also my production batch as well in the batch took full 24 hours in progress and it expired and returned nothing.
hope this helps: batch_6a99239e559c81908f1b5378c1382816
