Responses API Error: upstream connect error

API
,
1

I had a user complain of intermittent issues on 27 November 11:19AM. Viewing the logs we found this error being returned from a call to the Responses API (using GPT-5):

upstream connect error or disconnect/reset before headers. reset reason: connection termination

The error occurred 4 times. From a search on this forum it seems like an OpenAI server issue but no downtime was recorded on the API status page. Anyone else experience this and knows what could cause it (could it be something that I’m doing)?

1 Like
2

Correction - using GPT-5.1

3

Getting same error via api

1 Like
4

Received this error again this morning.

5

Hi!

API connectivity should always be assumed to be down. A retry system with exponential backoff delays should be implemented to retry failed connections along with a suitable mechanism to keep the end user informed of their current status.

With a globally distributed system there is a high likelihood of transient performance issues of one kind or another, that’s just the nature of a inter connected network.

1 Like
6

This sounds like the 503 error (The servers are currently overloaded)

This happens every now and then. No big deal. Just notify users and inform them to retry. Note: You need to correctly trap the 503 error.

no downtime was recorded on the API status page

The above notification is not necessary because the system was not down - it’s was just temporarily overloaded.

Here are the OpenAI error codes: https://platform.openai.com/docs/guides/error-codes

1 Like