API calls capped at 60 seconds (maybe only on GPT-4 Turbo)? (solved: whitelist openai on VPN)

I’m using gpt-4-1106-preview.

Starting around… December 4th, 5pm ET or so, I’ve been getting closed connections from OpenAI API. Some debug output indicates that requests are no longer allowed to take more than 60 seconds. This is tricky, since,well, I need the API to do the work in order to, well, work with it.

I am sending, if I count correctly (rust tiktoken library), about 1500 tokens, but that’s not changed from what I used to do.

Is this an intentional change on the OpenAI side? A Cloudflare configuration bug? An unintended config deploy side effect from a new SRE team member? Is my account somehow capped (that would surprise me, I am not a power user by any stretch of the imagination).

Welcome to the forum!

Sounds like it might be timing out on your end at 60 seconds? What error is it throwing? The new GPT-4 can take a while to reply for > 1000 tokens, even for me as a Tier 5 customer currently…

Hi Paul, thanks.

I have all the egg on my face. I enabled a VPN on my box a few days ago and apparently somehow had not had any requests longer than 60 seconds until ~30 hours ago, so I misdiagnosed.

Thanks for making me check again :slight_smile:

1 Like