Frequent 503 errors with Gemini-3-Flash-preview (~50% failure rate)

Gemini API
,
1

We are experiencing frequent 503 (Server Unavailable) errors while using the Gemini-3-Flash-preview model via Gemini API by creating genai google client.

Details:

  • Model: Gemini-3-Flash-preview

  • Error: 503 Server Unavailable

  • Frequency: ~50–70% of requests failing

  • Duration: Last 1–2 weeks

  • Usage Type: Batch processing and multi-row/ fields comparison

Observations:

  • Errors occur randomly even under normal load

  • Not exceeding RPM limits (verified in Google AI Studio)

  • Retry sometimes works, but not consistently

Impact:

  • Blocking our project workflows

  • Unable to reliably process data

Is there any known instability or overload issue with this model?
Any recommended workaround ?

2

Our team is actively working on a fix for Batch API issue where jobs are stuck in a pending state. If you are trying to use Batch API, I’d recommend waiting until the issue is resolved and then try again.