We are experiencing a problem since Friday last week. The GPT 3.5 calls (API) have become so slow, they are basically unusable.
Through some troubleshooting, we’ve managed to narrow down the issue to specific Orgs. One Org is exceptionally slow, while the other with the same parameter and identical prompt is substantially faster.
All tests have been done from the same machine and network as well as with identical settings. Even when testing the Playground, the responses in one Org are 300% slower on average compared the alternative one. There is no doubt that certain organizations are being throttled in some way.
This casts a huge shadow on OpenAI and the potential commercial applications of the GPT models. This is coming from one of the larger business groups in Europe, days before rolling out the first large-scale AI features.
The short-term “solution” is to use 2-3 fallback accounts from our subsidiaries. The long-term solution is to find reliable alternatives to OpenAI.
Hi and welcome to the Developer Forum!
This does seem to be something that a number of users are experiencing, you can report it to help.openai.com and use the support bot in the bottom right corner. Hopefully it gets resolved soon.
The chatbot was very unhelpful. And resulted in me writing a long bug report in the window and them saying they’ll get back to me “in about a week.”
I know it’s not a lot of money, but it is a paid product and the continued slowness without notice makes long term development with the product questionable.
I am adding on here because I am seeing this as well. Considering switching to bedrock
We’ve similarly experience a MASSIVE slow down for the past week or so. To the point it effectively breaks our product(s).
I have noticed it with users in the UK. A request takes roughly 20 seconds regardless of length whereas in the US it’s under half a second. Same request.
Same here; it is insanely slow and barely usable. Looking for other options. Really disapointed with both the Chat and the API, the overall quality is going downhill!
I have one user who was running some requests that were about 150 tokens total and each request was taking on average 20 seconds for OpenAI to respond. He is in the UK. On my end though, I’m in the US and it only took half a second for OpenAI to respond in each instance. We were completely stumped so I had to log the timecodes in a debug file (below). Something weird and messed up is going on at OpenAI.
We’ve experience a massive slow down for the past week.
Any update ?
Consider starting a fresh post for your query.
Typically, older topics get closed when brought back up.