Hi all, we’re seeing a recent and significant rise in timeouts on the OpenAI Responses API, and want to check if others are seeing the same.
We’ve been running this flow in production for several months with no major timeout issues. The spike appears to have started recently (mid-April 2026).
Setup
Mobile app calls a Supabase Edge Function
Uses:
POST /v1/responses
model: gpt-4.1-mini
Timeout:
We allow 45s for a response before aborting it as a timeout
What changed
We’re seeing many more timeout failures. These calls have fairly consistent token usage and a typical average latency around 10 seconds, so 45s+ responses are unexpected.
Observed timeout rates (CT)
2026-04-16: 2 / 156 = 1.28%
2026-04-17: 3 / 176 = 1.70%
2026-04-18: 7 / 185 = 3.78%
2026-04-19 (partial day): 9 / 66 = 13.64%
Latest 24h vs previous 24h:
7.29% vs 1.88%
Is anyone else seeing a recent increase in gpt-4.1-mini / Responses API latency tails or timeout behavior in the last 1–2 weeks?
Thanks for the response, Mark. Things do seem to be better now.
You mention this lines up with the recent Responses API outage, but looking at the status page, I see a “Responses API Streaming Error” listed for the 16th, but the 17th, 18th, and 19th, which were the worst days for us, report “No Incidents”.
Is there any more reliable way to determine if this is an outage, in the future?
Hey @lastround360, yeah, that mismatch between what you experienced and the status page is pretty confusing.
What’s going on is the status page only reflects incidents once they’re confirmed and scoped by the engineering team. So if something is intermittent, region-specific, or still being investigated, it might not show up at all or only appear later. That’s why those “No Incidents” days can still feel rough on your side.
Right now there isn’t a super reliable real-time signal to confirm “this is definitely an outage” from the outside. It often ends up being a process of ruling things out locally first.
A few things that have helped others:
Check if errors are consistent across different endpoints or models
Try from another region or environment if possible
Watch community threads to see if others are reporting the same pattern
If it’s still unclear, opening a ticket at support@openai.com is the best way to get it looked at more deeply with logs. That’s usually how these gaps get confirmed and later reflected (or improved) on the status page.
This isn’t ideal, agreed. Better visibility here would go a long way.
Having been on this forum for a few years, a clear outage—such as being unable to connect—usually becomes obvious quickly. The forum tends to fill with “me too” replies within minutes.
For more subtle issues like the one you described, starting a topic—as you’ve done—is one of the most effective ways to gain visibility. Discourse threads are indexed by search engines very quickly, and they often attract others experiencing the same problem.
If the issue persists, it can also help to bump the thread periodically. Ongoing activity increases visibility and signals that the problem remains unresolved.