Persistent 503 Errors on Gemini Batch API GET Endpoint — Jobs Succeeded but Results Unretrievable

Gemini API
,
1

I’m encountering a critical issue with the Google Generative AI Batch API where batch jobs complete successfully, but the GET batch endpoint consistently returns 503, making it impossible to retrieve the results.

Summary

  • Service: Generative AI Batch API

  • Endpoint: generativelanguage.googleapis.com/v1beta/batches/{jobName}

  • Error: 503 Service Unavailable

  • Duration: Ongoing for ~2+ hours

  • Impact: 747 successfully generated images are inaccessible

API Behavior

:cross_mark: GET /v1beta/batches/{jobName}

Returns 503 Service Unavailable consistently.

Example:

{
  "error": {
    "code": 503,
    "message": "The service is currently unavailable.",
    "status": "UNAVAILABLE"
  }
}

This endpoint is required to access:

  • metadata.output.inlinedResponses

  • OR destUri / output file references

:white_check_mark: GET /v1beta/batches

  • Works correctly

  • Returns job metadata (state, stats)

  • Does not include output results

What I’ve Tried

  • Repeated retries with exponential backoff

  • Different endpoint paths (v1beta, v1)

  • Verified API key validity (other endpoints work)

  • Confirmed jobs are SUCCEEDED

  • Checked Google Cloud Status Dashboard (no incident reported)

Expected Behavior

The batch GET endpoint should return completed batch details including:

  • Inline responses or

  • Output file location (destUri)

Actual Behavior

The endpoint returns 503 indefinitely, blocking access to completed batch results.

  1. Is this a known issue with the Gemini Batch API?

  2. Are completed batch results recoverable once the endpoint is restored?

  3. Is this region-specific or a broader outage?

  4. Should we wait, or are results at risk of being lost?

  5. Any recommended workaround for retrieving outputs

3

Tried to do another batch 17 hours later, seems to generate the images but still throws a 503 when retrieving the results.

Constant 503 error when using Batch API for video processing
4

I was able to run a batch of 10 requests successfully just now. The failing batches were holding 500 and 100 requests each.

6

@Sonali_Kumari1 can you help?