1 min+ for assistant API to answer

API Feedback
22

We can’t have enough topics from this! I’m currently not using it for something that critical but it is very annoying and already two days indeed. I contacted OpenAI three times. No decent answers to work with. I asked my bot what to and it give me an answer that the people at OpenAI could use probably and their support will see the irony a gpt generated advise:

  1. CPU Usage: Check if the CPU is being heavily utilized.
  2. Memory Usage: See if there’s enough RAM available.
  3. Disk I/O: Determine if there’s high disk usage or slow read/write speeds.
  4. Network Latency: If the AI agents are network-dependent, check for network issues.
  5. Running Processes: Identify if there are any processes consuming excessive resources.
3 Likes
23

This should be classed as a CRITICAL OUTAGE, what’s going on, can we get some answers!?

5 Likes
24

My application is basically down because of this and unusuable, but downtime isn’t even registered on OpenAI’s status page. Issues like this make me consider migrating off of the Assistant API.

4 Likes
25

This issue gets better now, I guess it’s the server load problem of OpenAI?

2 Likes
26

Same for me it looks like it works again !

2 Likes
27

Still getting extremely slow responses

3 Likes
28

Me too unfortunately. Really troublesome atm. Hopefully the OPEN AI team fixes this soon.

2 Likes
29

It’s back to normal for me now. Which is still slow but usable.

UPDATE: It’s terrible again. Borderline unusuable.

2 Likes
30

For me its really really slow. 15 second answers. Three weeks ago it was 6 seconds.

2 Likes
31

For me, it takes 45 seconds just to answer ’ Hello’.

5 Likes
33

Mine are still slow, like very slow.
Im using an Assistant (o1) with file search on an attached vector database.
The data is small and well organized.

It takes 184 seconds to give back the answer

1 Like
34

That’s less surprising, it’s a reasoning model.

1 Like
35

I thought using o1 because the accuracy of gpt 4 was not good enough even for simple prompts. It has a vector database attached with the information of a list of Yachts.

It should answer back with the ID of the yacth. But it fails even when asked simple things like:
Show me boats longer than 50 in lenght

With Gpt 4o the response time is 48 secs for a simple prompt like that and it gets it wrong

Which approach should i use for this type of use case?

I had better results when I was passing the yacht list as the API response from my database. But in that moment there weren’t so many fields in the table and neither lot of data

As an alternative, Im exploring to use a text generation for creating the actual filters of my API Call that returns the yacht list

2 Likes
36

Yeah this doesn’t make sense. That’s waaaaay too long.

37

I can’t even get a valid response in time before my Lambda execution timeout. This is unbelievable.

2 Likes
38

This was my fault :sweat_smile:
I made a new JSON object containing only the fields that AI could use to search and now the response time is around 7 secs but quite accurate :+1:

2 Likes
39

Sorry for the issues — we’ve made some adjustments here (last Tues night/Weds morning) to speed things up and it looks like responses are faster now, at least from our end. Are things better for you?

1 Like
40

The 1 minute response times are gone since a few days back, but we still have frequent 15sec+ answers, with the exact same set-up we achieved 6-8sec answers about 1 month ago. So this should still be classed as a CRITICAL OUTAGE

41

Do you know anything related to attachments in a message to Assistant API? a week ago the response times I had when attaching a PDF document to a message and requesting a function calling was around 10-30 seconds. Currently, this is taking over a minute and a half. Has there been any change in this regard?

42

I can only agree with Niklas. For me the issue still persists. The answers are still to slow.