The OpenAI support account doesn’t log in like a normal user - it is a collective account that uses the forum’s backend API or similar, and doesn’t get pings when you tag it.
The performance issues with the Assistants API has been flagged to OpenAI with the channel available to communicate with them internally from the forum, and we can only wait for it to be recognized on status.openai.com and investigated. It is 7:30am in San Francisco.
problem here in Italy… news?
We’ve been experiencing serious issues with the Assistants API since this morning (7:54 CET, Germany) aswell.
No responses are returned from our endpoints — requests just hang or time out.
Even worse, the usage dashboard continues to count tokens (input, cached input and output) despite no successful responses.
We’ve temporarily taken our service offline until the issue is resolved.
Had to put our service offline. When will this be fixed? Our income is dependant on this and we are leaving a bad impression on our customer. Can you provide info Open AI?@OpenAI_Support
This is a mess, we’ll have to create a switch to change to Claude or Gemini in these cases.
A huge mess I agree. This cannot just randomly happen to a system we are paying for and are dependant upon. Will be considering a switch to another provider
When will this problem be resolved? Any updates from support? @OpenAI_Support
Same here. Assistants Api showing logged thread in the left hand side but when i choose the thread i first show nothing then after a few minuets the page loads data but no response output shows at all.
Historically the assistants API has been solid, but seems adding the agent builder to the playground has caused a bunch of issues.
Might be a good impetus to get moving in the migration to Responses:
https://platform.openai.com/docs/assistants/migration
There is no easy path where you can simply run ‘migration tool’ to port over things like existing threads, but you can consider getting your rollout going on Responses (where you pay for those file searches).
Then increase your resilience against continuing API adversity by keeping embeddings-based document search and conversation context in-house.
I’ll add: Responses also isn’t currently desirable, because there is no budget constraint on the amount of conversation run each turn…
I am also having the same problems being reported by others. I am in Germany.
Same here @OpenAI_Support At least some information would be highly appreciated. We have this problem since yesterday. And as it seems it is still an issue. We are from Germany. Funny thing. We wanted to deploy an app tomorrow. Now the client is getting nervous.
Hi everyone,
I’m also experiencing significant slowdowns when using the API (I’m based in France).
It’s quite disappointing to see that issues like this have been happening fairly often since I started using the API in July.
It would be very helpful if such incidents were reflected on the status page (https://status.openai.com/), so that developers could at least communicate more transparently with their own clients.
Good luck to everyone working hard to get this resolved 
Same issues here in California. Taking way too long to respond, timing out, hitting errors, basically breaking our app @OpenAI_Support Please give update.
Los mismos problemas desde Uruguay, ya estamos viendo de usar Gemini.
same problem here, api responses very slow
Same issue in UK. API Assistants slow or non responsive resulting in existing processes not working. I tried enabling the ‘Priority’ function on the project to see if that would result in a quicker response but no identifiable change to performance, just a higher cost for the liberty of sending the request. I’ve also been notified that the platform we use has had other users raise the same.
The issue has now been acknowledged by OpenAI.
According to the official status page (Link to the Incident), the team is aware of degraded performance with the Assistants API and is currently investigating.
Same here in Finland. We had an academic research data collection set up for today, with recruited research participants being supposed to use our AI agent. This was ruined by the outage.
