Repository navigation
Gemini Batch API jobs with gemini-3.1-flash-preview do not reach SUCCEEDED after 24+ hours #2974
Description
Description
Since approximately 2026-09-15, Gemini Developer API (AI Studio API) batch jobs using gemini-3.1-flash-preview have not transitioned to JOB_STATE_SUCCEEDED, even after more than 24 hours.
The jobs are created successfully, but polling them with client.batches.get() continues to return a non-terminal state without a useful error. This started suddenly in a production catalog-processing pipeline; the same integration had previously completed batch jobs normally.
This appears similar to past incidents reported in:
- Gemini Batch API stuck all jobs on BATCH_STATE_PENDING for more than 24 hours #1482
- Gemini 2.5 Pro Batch API: Jobs stuck in PENDING/RUNNING for over 24 hours #2221
Both previous issues are closed, so I am reporting this as a possible recurrence.
Environment
- API: Gemini Developer API / Google AI Studio API (not Vertex AI)
- Language: Python
- SDK:
google-genai==1.60.0 - Model:
models/gemini-3.1-flash-preview - Input: JSONL file uploaded through the Files API
- First observed: 2026-09-15
Minimal flow
import os
from google import genai
client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
uploaded_file = client.files.upload(
file="input.jsonl",
config={"mime_type": "jsonl"},
)
job = client.batches.create(
model="models/gemini-3.1-flash-preview",
src=uploaded_file.name,
config={"display_name": "batch-test"},
)
# Still does not reach JOB_STATE_SUCCEEDED after 24+ hours
current_job = client.batches.get(name=job.name)
print(current_job.state)Expected behavior
The batch job should complete within the documented processing window, or fail with an actionable error.
Actual behavior
The job remains in a non-terminal state for more than 24 hours and does not produce output or an actionable error.
Additional information
We can provide affected batch job names privately to maintainers if needed, but have omitted account/project identifiers and job IDs from this public report.
Could you confirm whether there is a current service-side incident or elevated queueing for Gemini Developer API batch jobs using gemini-3.1-flash-preview?
Activity
grhan17 commentedon Sep 16, 2026
Additional reports from the Google AI Developers Forum suggest this may be a broader Batch API incident during the same time window:
gemini-3.1-flash-imageandgemini-3.6-flash, beginning around the evening of September 14.gemini-3.5-flash-lite. The affected jobs eventually reachedSUCCEEDED, but all output entries containedInternal error encountered. The reporter later noted that newly submitted jobs had started processing normally again.The models are not identical to ours, but the timing and Batch API behavior are closely aligned: jobs are accepted, remain in
BATCH_STATE_RUNNINGwith no processing for many hours, while non-batch requests continue to work. This suggests a possible service-side Batch scheduling or processing incident rather than an SDK-specific issue.Venkaiahbabuneelam commentedon Sep 16, 2026
Hi @grhan17, Thank you for bringing this to our attention.
We have tested and confirmed that stable models (such as gemini-3.5-flash and gemini-3.5-flash-lite) complete successfully without issues.
Thanks
GinaMhd commentedon Sep 22, 2026
I have the same problem for the last 16 hours now, the batch requests are being accepted, but they remain stuck. Im using gemini-3.1-flash-lite.
amartynyuk commentedon Sep 25, 2026
Same issue with gemini-3.8-flash using global endpoint.
Venkaiahbabuneelam commentedon Sep 28, 2026
Thank you for bringing this to our attention. We truly appreciate you flagging this issue, we will file a bug internally.
Venkaiahbabuneelam commentedon Sep 29, 2026
Hi Team,
The issue has been resolved by the latest SDK. We confirmed that jobs are being received successfully via gemini-3.8-flash and gemini-3.1-flash-lite. Please check and let me know if the issue persists.
Thanks
amartynyuk commentedon Sep 29, 2026
Confirming, jobs now succeed on gemini-3.8-flash global. Thank you!
Venkaiahbabuneelam commentedon Sep 29, 2026
Thank you for the confirmation. I am now closing the issue, if it persists, Kindly please create a new one.