Gemini Flash is suddenly showing overload / DECODE_PREEMPTED errors

Gemini API
1

I am developing an Android application called Substation AI Pro using Firebase AI and Gemini Flash.

Recently, Gemini Flash has started returning repeated overload errors during image analysis, although the same application and workflow were working successfully before.

Firebase AI Monitoring shows errors such as:

UNAVAILABLE
DECODE_PREEMPTED
“Preempted out of decode queue by a higher priority request”
INTERNAL
“This model is currently experiencing high demand.”

I have tested multiple circuit images, and the problem is not limited to one particular image. I also observed similar behavior with another Gemini Flash model.

Please clarify:

  1. Why has this overload problem started recently when the same application was working normally before?
  2. Is there currently any known Gemini Flash backend capacity, regional routing, or service issue affecting users?
  3. Do failed requests with UNAVAILABLE / DECODE_PREEMPTED count against the Free Tier daily request quota?
  4. Is this a temporary Google-side problem, or is there anything I need to change in my Android/Firebase implementation?

Please investigate and advise.

2

Hello @Rajesh_Mangalapuram ,

To help us investigate these errors further, could you share your Project ID, the exact Gemini Flash model string you are calling, and a sample prompt via direct message?

3

Hello @Rajesh_Mangalapuram ,

Thanks for sharing the details. These errors mean the model was under high demand at the time. Your project is on the Free tier. For a production app, we recommend setting up billing to move to a paid tier and retrying failed requests with exponential backoff: Handling 429/503 errors.