Background mode requests stuck in ‘in_progress’ for 30+ minutes - Responses API

API Bugs
,
1

Hi @OpenAI_Support, this seems to be a repeat of a previous bug [“Background mode requests stuck in ‘queued’ forever - Responses API”] which seems to crop up from time to time (just browse the “background-mode” tag).

For the last few days, my API calls using the Responses API and background mode have been failing intermittently with no rhyme or reason. I haven’t changed anything in my API setup. This is with GPT 5.6 sol and luna. The OpenAI dashboard shows “No output items.” The status of the call will be stuck in “in_progress” for 30+ minutes when the call takes 2-3 minutes max using the OpenAI Playground/Chat with the same exact prompt and reasoning effort.

Please take a look at this urgently as my app’s users are complaining and there’s literally nothing I can do about this other than hope it gets fixed (as it did previously). Thank you!

Edit/Update:

Some additional information:- The issue only happens with the 5.6 family of models, I can’t replicate this using 5.2, 5.4, or o3.

  • Again, this only happens with background mode, normal synchronous GPT 5.6 calls are not affected.

There is definitely something specific with the combination of:

  • background mode
  • GPT 5.6 luna/sol (haven’t tested terra)
  • higher reasoning efforts (medium+ on Sol, max effort on luna)

To recap, an API call that normally takes 2-3 minutes is taking 30-60 minutes to exit the “in_progress” state.

Edit/Update #2:

This seems to have started around the Luna price cuts and “inference optimizations” that introduced “Fast mode” (previously Priority tier).

2

Hi and welcome to the community!

Can you share a few request IDs that are or were stuck in progress?

I just fired off 100 background requests using Responses with Sol and Luna, but they all came back as expected.

Is there anything else you can share for a clean repro?

4

Some additional information:

  • The issue only happens with the 5.6 family of models, I can’t replicate this using 5.2, 5.4, or o3.

- Again, this only happens with background mode, normal synchronous GPT 5.6 calls are not affected.

There is definitely something specific with the combination of:

  • background mode
  • GPT 5.6 luna/sol (haven’t tested terra)
  • higher reasoning efforts (medium+ on Sol, max effort on luna)

To recap, an API call that normally takes 2-3 minutes is taking 30-60 minutes to exit the “in_progress” state.

5

I can’t be the only one experiencing this, if you’re dealing with the same problem please create a forum account and chime in!!

Also, hit the “Me too” button in the OP :slight_smile:

6

Do you have a request ID for an API request using background mode that is or was stuck in progress? This would be the most helpful information when forwarding the issue to the team.

7

Yes, I sent a bunch of request IDs in my support ticket.

8

It seems they silently fixed this with no acknowledgement. I still didn’t change anything, it just started working again and I can’t reproduce the high-latency calls anymore. Really annoying how support has no knowledge of what engineering is doing or vice versa. Billions of dollars yet operates like a small startup. I guess I should be happy they fixed it.