Gemini 3.1 Pro Batch jobs stuck with zero requests processed across two Tier 2 projects

Gemini API
,
1

Seven Gemini Developer API Batch jobs using gemini-3.1-pro-preview remain stuck in JOB_STATE_RUNNING with zero requests processed.

The issue occurs across two separate Tier 2 projects with different API keys. A normal generateContent request works immediately, while even a Batch containing only one request remains stuck without an error.

The Batch quota dashboard shows usage far below the limits: 6/100 concurrent Batch requests and 2.48M/500M enqueued tokens.

Cancellation was requested for five of the jobs, but they have not reached a terminal CANCELLED state.

I have already submitted a detailed private report through Google AI Studio. Full project numbers, operation IDs and sanitized REST responses are available privately to verified Google staff.

Is there a current Batch API incident, or could a Google team member investigate this?

2

im having the same issue rn i’ve already sent several small test jobs they all stuck at batch_state_running..

3

Same here with gemini-3.1-flash-image via Batch API. Since the evening of September 14, every new job stays in BATCH_STATE_RUNNING with all requests pending and updateTime equal to createTime, including a job with a single request. An identical job on the morning of September 14 finished in about 5 minutes; nothing changed on our side.

4

Yeah, stuck here since before midnight EST yesterday. This is the 2nd time recently this happened and it’s painful.

We can run live requests but not batch, but our workflow depends on lots of tiny requests so we blow through the rate limit in no time running live and then we’re offline til midnight PST.

In fact the only reason we use batch is because of rate limit, not cost. Stupid really.

Let’s hope they fix this soon, we’ve lost most of the day today and we’re starting to get antsy.

5

Seeing the same problem since yesterday as well with gemini-3.6-flash with the batch API, all stuck on BATCH_STATE_RUNNING…

6

I just got one out after running nearly 17 hours, still got another 5 pending but something is happening at least. Not a total stop, but still…at this rate what we do in 3-6 hours could take a week.

7

same here from 9/15. more than 24 hours. I’m using gemini-3.1-flash-preview batch API w/ python

8

It’s back again (in case anyone is subscribed to this thread). But check your results, I had one set that was completely blank and needed a rerun.

The others seem fine and its back.

9

Thank you. Working for me too now.

10

We’ve been stuck since Friday and are still stuck. Anyone else still stuck?

11

In my case, the original stuck jobs eventually completed, but all 859 requests failed with an internal error. After the service recovered, a five-request test succeeded, and resubmitting the remaining requests in smaller batches worked correctly.