ALL of The Gemini Models Are giving me 503 Error

Gemini API
, ,
1

Is this a global outage or what? I’ve tried Gemini flash 2.5, gemini 2.5 pro

all are rate limited or out of service (503)

although I’m tier 1 paid

1 Like
2

Yes it has been there for 2 weeks and nobody is fixing it. I tried to contact Google Gemini dev on X, this forum, nobody care apparently. Apps in prod are breaking with almost 50% 503 erros now.

3

I wish it was only 50%, at rush hours, it’s 100%, even switching models didn’t help leading to total failure of multiple apps i’ve had, it costs a lot to compensate users.

4

Yep im hitting 100% 503 error juste now. GOOGLE PLEASE FIX YOUR “Modified by moderator”

5

I guess is because of nano banana? cause this issue is happened at it released. Or computing resource is borrow to gemini3?

6

Yes, I believe that’s the cause, and it’s only going to get worse, since high-ranking people in Google’s AI department just recently boasted on X that Gemini is number one in mobile store downloads. All while ignoring the fact that the API has been unusable and breaking production apps for 2 weeks, it’s really disappointing.

2 Likes
8

Hello,

Welcome to the Forum,

A 503 error is often transient. As initial steps, we recommend implementing exponential backoff and retrying the request. You could also try switching models to check if the issue is model-specific.

If the problem persists, we would request a few additional details, such as:

  1. The timeframe when you observed the errors
  2. Whether it occurs with a specific model or multiple models
  3. The frequency of these errors

This information will help us analyze your issue more effectively and provide more accurate assistance.

10

plz fix it I’ve encountered the same situation that can’t fix it at our end

11

I wouldn’t even be here if these so-called “fixes” actually fixed anything.

I’ve tried it all. retries with jitter, exponential backoffs, every programmable safeguard on my side, even falling back to older, clunkier models.

And yet, every single time between 2–8 PM GMT+3, 2.5 Pro just dies on me with endless 503s. The fallback to Gemini 2.5 Flash isn’t any better either, it throws 429s on the very first request, completely unjustified.

It’s honestly a mess. I’ve even had to drop down to Flash Lite just to get something through, and even then it spat back 429s nonstop.

Pretty bad. Really bad. Unusable.

1 Like
13

Hello,

We sincerely apologize for the trouble you have experienced, and thank you for sharing the initial details. May I kindly request you to provide the complete error details for the 429 error, along with a snippet of your code and some example prompts where you are encountering this issue? This information will help me reproduce the scenario and prepare a detailed analysis report for the concerned team.

Thank you for your patience and cooperation.

15

Hello Lalit_Kumar,

Thank you for requesting more details. iI’ve been getting the same error.

I have specific findings related to the 503 and subsequent 429 errors that I believe point to an issue within the Google GenAI Python Library.

My setup uses the following configuration:

  • Library: google-genai (Python SDK)

  • Configuration: genai.configure(api_key=self._api_key).

    Model Selection: self._model = genai.GenerativeModel(config.model_name) (currently Gemini 2.5 Flash).

  • Execution: response = self._model.generate_content(prompt.contents).

My analysis suggests that the persistent 429 errors are a consequence of the automatic retry mechanism embedded in the GenAI Python Library, which is triggered by the initial 503 errors.

When a 503 occurs, the library’s retry logic executes extensively, leading to a significant and often unexpected increase in the number of API calls made against my quota.

Specific Observation:

I had a workload requiring only 25 intended requests. Due to repeated 503 errors, the underlying library generated an estimated 250 attempts (retries). This action immediately consumed my daily Free Tier quota (250 uses/day).

The subsequent API calls then resulted in the wave of 429 errors, as the quota had already been exhausted by the internal retry mechanism. The 429 errors are thus a secondary effect of the GenAI Python Library’s retry behavior following service instability (503), not an issue with my original request rate.

Thank you for your time and assistance.