Gemini batch API 出现503错误，模型过载

Gemini API
,
1

Hello Google Team,

I am on a paid Tier-3 plan, but I keep encountering issues when using the Batch API for video understanding.
I am using the Gemini-2.5-Pro model. Each video I upload is 134 MB in size and about 10 minutes long, and my prompt is roughly 700 characters.

Every run returns a 503 Service Unavailable error:
“Error: Model is overloaded. Please try again later.”

3

Hi @Zoe1 ,

A 503 error (UNAVAILABLE) indicates that the service is temporarily overloaded or experiencing a capacity constraint. This is common during peak usage hours, particularly with long context requests. If you’re using code, implementing a retry with exponential backoff is the most robust way to handle this.

Could you try switching to a different model and see if that works? Additionally, what is the mediaResolution currently set to?

5

Hello,

Although the code already implements exponential backoff retry, we still receive 503 errors at runtime.
We switched to the Gemini 2.5 Flash model, but its output quality is noticeably lower than Gemini 2.5 Pro.
The mediaResolution is 1280×720.

e13e9b86d83d4b56b4bc9e32c8a27d31
e13e9b86d83d4b56b4bc9e32c8a27d311072×896 52.5 KB