Model is overloaded -

I have been receiving this error off and on since yesterday. I am not finding any notices about technical issues with Gemini. Any ideas?
Error: GoogleGenerativeAIFetchError: [GoogleGenerativeAI Error]: Error fetching from https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-002:generateContent: [503 Service Unavailable] The model is overloaded. Please try again later.

3 Likes

Hi @Ramsar, Welcome to the forum !!!

We see this error sometimes when there are so many requests or service is down. I recommend temporarily switching to another model and see if it works. Or wait a bit and retry your request. It will resolve itself.

Thanks.

1 Like

We see this in one in every 6 requests or so, which is new. Further - we have a fallback to gemini pro when that happens and it STILL is overloaded.

1 Like

Hey everyone,

I’ve been facing 503 errors very frequently, and it’s getting extremely frustrating. Almost 90% of my requests are failing with a 503, making Gemini practically unusable for me.

I even tried switching models, but the issue persists. Does anyone else have a fix or workaround? Would really appreciate any help here!

3 Likes

Yup even I tried to change models . Still the issue is same

1 Like

I’m also getting this error on most of my requests. It’s been completely unusable for the last 24h - has the issue even been acknowledged by Google? I wonder if they are allocating servers away from 1.5 and into 2.0 due to an imminent launch. Even if that’s the case, it sucks for us, their current paying customers.

1 Like

Has anyone come up with a workaround or resolution for this yet? We’re seeing the same thing in our apps. Greater than 80% of requests are failing with the 503 Model is Overloaded error regardless of the Gemini model we’re using.

1 Like

Hey folks, we are not moving capacity away from 1.5 models right now. Looks like we might be getting hit by a DDoS attack. Will follow up as we mitigate this.

6 Likes

I have the same problem. Lots of errors today ā€œThe model is overloaded. Please try again later.ā€. Using gemini-1.5-flash-8b, gemini-2.0-flash-exp or gemini-1.5-pro makes no difference.

1 Like

So the Hebrew input and output interrupt caused by it?plz check it.

1 Like

How can we use this API in production in our apps when almost 70% of requests result in this overloaded kind of error?

4 Likes

For me too, more than 50% of requests were in error last night. Can we have a real answer please? Otherwise, I’ll have to change service provider…

2 Likes

What scares me the most is the lack of acknowledgement of how bad it is beyond Logan’s earlier reply in this thread (no blog posts, emails to customers, a plan to fix it, a timeline…).

2 Likes

Around this time we saw this happen to one in every 6 requests or so across many days. We even had it switch to G 1.5 pro and it was ALSO overloaded. Why is there no status reports for this? Similar to when pro went functionally down for over 6 hours, with 25second+ response times per query - this didn’t show up on ANY alerts or status boards I could see.

@Logan_Kilpatrick any place we can get more info, or more realtime status of gemini failures? It took us a while last time to determine it was gemini (we just didn’t expect it) we have better monitoring in place now, but it’d be good to know when ā€œit’s overā€

4 Likes

Getting the same error generateContent: [503 Service Unavailable] The model is overloaded. Please try again later.

Can’t find realtime api status anywhere?

1 Like

Same here. Driving me insane

I’ve built an entire app that relies on this api, when it goes down like this, my app breaksšŸ™ˆ, and I can’t use any other provider as they all token and cost limited, gemini is uncapped(in theory), but this 503 error has been happening too much.

1 Like

Jamie, this is what we’ve done - and I’d recommend having a fallback of another of their models that seems to be on different infrastructure, such as Gemini 2.0 flash - lite, it has the same cost as Gemini 1.5 flash 2.0.

Just tried that, did not work. Tried every model. They all are throwing 503.

This issue has been happening throughout the entire past 2 days. Why is their API under so much strain to the point where it is completely inoperable?
Are they under a DDOS again?