1

Hi, my GPT-4 API calls have been extremely slow since today. I am experiencing the same issue with ChatGPT (GPT-4). It seems to be linked to my account because when I test the speed using another account, GPT-4 responds normally. I havenā€™t reached the maximum quota for this month. Could it be related to a quota issue? Are there any other possible reasons?

Thanks for your help

2

The same here. Timeout every times after 5min. I need to launch 5 times to get 1 and i pay for 5, really amazing ! What is your plan to solve that OPENAI ?

1 Like
3

Yup, this is happening to me as well. Also seems to be specifically connected to my account.

1 Like
4

They have issues in their servers, I donā€™t work for OpenAI but it usually happens when they release new features, and plugins probably got everyone excited and got jumping on, which uses GPT4 in the background.

I hope they solve it soon.

1 Like
5

Yes, it was happening to me also. So I changed my backend response to SSE (Server send events) so that my users should know that something is going on.

1 Like
6

Overall average today for GPT-4 is 30s, but a few spikes around 6 minutes:

Screenshot 2023-05-17 at 11.17.29 AM
Screenshot 2023-05-17 at 11.17.29 AM1074Ɨ832 54.2 KB

3 Likes
7

Impressive! May I ask where you obtained this data from?

1 Like
8

This is from the monitor in AWS hooked to the Lambda function that makes calls to the GPT-4 API.

So my own personal view of performance from my API calls.

1 Like
9

Hereā€™s our API performance for GPT-3.5-turbo over the last month:

image
image2099Ɨ779 60 KB

The data are somewhat noisy, but clearly an upwards trend. The type of inference hasnā€™t changed and the size of each request hasnā€™t changed.

2 Likes
10

I wonder what the solution is, especially if you sre trying to build a business around it.

11

Well, itā€™s not clear that there is a solution yet.
Iā€™m hoping the OpenAI folks will pull through.
Iā€™d be OK paying, say, 10x more for a 4x faster product with some kind of latency guarantee.

Iā€™m also looking at other models, both self-hosted, and competing providers, but itā€™s somewhat slim pickings for now ā€¦ Especially until Google actually lets people use that PaLM model they ā€œreleasedā€ twice already.

Clearly, thereā€™s demand for what OpenAI are doing, and the main challenge is meeting up to that demand.

1 Like