Hi Mustan. I am also facing same issue. Is it resolved?
Hi
@Amruta_Hedge could you share more details
- What model are you using ?
- Are all your request returning a 503 error ?
Hello Mustan. Thank you for the response. I am using gemini-3.1-flash-lite-preview. Not for all requests. 503 error came for first time then on re request it resolved. But why it came I am not sure. we have used it in one of our projects. So such errors should not come in production application. So I am worried as we cant control such errors
Hi @Amruta_Hedge , 503 errors are due to capacity and not related to your quote,I do recommend implementing an exponential backoff and retry (our SDKs support this through retry config)
I just started getting this issue. Started from yesterday morning randomly and today 100% getting 503. I’m using the gemini 2.5 flash model.
1 Like
Hi! I’ve been developing with Gemini 2.5 Flash (free tier) for 3–4 weeks with zero issues.
Right before rollout, I enabled billing, and immediately started getting 503 errors on about 90% of requests (now nearly 100%).
Nothing else changed:
same request structure
very small batches (even 1–5 items fail)
low volume, cron jobs spaced out and not concurrent
Now even simple test calls fail consistently for hours.
2.5-lite works, but quality is not sufficient.
Everything worked fine until billing was enabled. The only other change was scheduling 3 single small cron jobs the night before.
Before the 503s, I briefly started getting request denials referencing “automated scripts” (which is odd for an API). After adjusting the curl request, calls started going through, but shortly after, the 503 errors began and are now nearly constant.
I’m also seeing many reports of persistent 503 issues, so my question are:
Can enabling billing or running scheduled cron jobs (low volume, spaced) trigger this?
Has anyone successfully stabilized 2.5 Flash in production?
Is there any real fix, or is switching providers an easier solution?
Additional observation:
As an additional test, I temporary switched to the free-tier API key used for testing, and the exact same requests work without issues. Switching back to the production billing-enabled key immediately brings back the 503 errors.
Since yesterday, almost ALL of our calls are returning 503. It’s not just one or another, and it’s been happening throughout the WHOLE day. There should be an API status page so we can understand what is happening and if we can switch to another model temporarily.
It doesn’t help if all the retries also return 503 and you get charged even more for requests that weren’t completed
@faisst for which model are you experiencing this for ?
@Mustan_lokhand Hi Mustan, I am using “gemini-3-flash-preview” for OCR task, and I am getting this 503 error. I tried with a different API, but I am getting the same error. I wanna know how this works? Is it that my request/quota got over, or the model is too busy on Google’s server to handle the new request? Will Vertex AI solve my problem?
Even I am experiencing 503 error with 2.5 pro and 3.1 pro (gemini api keys) is silently failing…Performance of my chatbot is severely affected due to this..error rate has gone above 60%
2 Likes
Same issue here. 503 with gemini 2.5 pro for more than 4 hours.
2 Likes
Started with Gemini 2.5 Pro and Flash, tried to switch for 3.1 Pro but the issue remained
1 Like
Same issue here.
Just as i was about to launch my project yesterday i started getting 429 error in the terminal and 503 error in the aistudio log.
Pro user, nowhere close to the limits.
Gemini 2.5 Flash API
1 Like
Any update on this regard?
The official page ( Google Cloud Service Health and Google AI Studio) reports no problems at all when users are reporting 40% error rate in their API calls. That is just unacceptable.