Help escalating Gemini Batch API issue

[dani-mp]

Hi,

We’re running into an issue in production with Gemini batch jobs and I’m hoping you can help connect us with the right team.

Since last Thursday, new Gemini batch jobs that we create have been getting stuck in PENDING for more than 24 hours. Before that, these same jobs were usually completing within a few hours, even when some items failed and had to be retried.

What we’ve checked so far:

• the issue seems specific to Gemini batch jobs
• non-batch requests still go through
• billing/account settings look fine on our side
• we rotated the production Gemini API key in case the key had some issue, but that did not help
• we also updated our tooling to track batch creation/update times and to recreate stuck batches, but new batches are still getting stuck in PENDING

At this point it’s not clear to us whether this is:

• a service-side issue
• an account/project-specific issue
• a quota/queueing problem
• or something else

This is a significant issue for us operationally, so we’d really appreciate being put in contact with someone on the appropriate Google/Gemini side who can help investigate.

Thanks.

[Varun-S10]

Hi @dani-mp, To help us investigate, please clarify if you are using Vertex AI or the Gemini API, and which version of the @google/genai you are on and which Gemini model you are targeting. Could you also share minimal reproducible code?

[dani-mp]

Hi, thanks for following up.

We’re using:

• @google/genai: ^1.46.0
• API path: Gemini API with a Gemini API key, not Vertex AI
• model: gemini-2.5-pro

Minimal repro:

import { GoogleGenAI } from '@google/genai'

const ai = new GoogleGenAI({
  apiKey: process.env.GEMINI_API_KEY!,
})

await ai.batches.create({
  model: 'gemini-2.5-pro',
  src: [...],
})

The issue on our side is that new batch jobs remain in JOB_STATE_PENDING for 24h+ and do not progress, while non-batch requests continue to work normally.

[Varun-S10]

Hi @dani-mp, Thank you for bringing this to our attention. We truly appreciate you flagging this issue, we will file a bug internally.

[dani-mp]

Hi @dani-mp, Thank you for bringing this to our attention. We truly appreciate you flagging this issue, we will file a bug internally.

Thanks. Note that I'm stuck with this issue since last Thursday, which is impacting our production systems. A whole week where the API creates batches that never complete.

[speedstorm1]

Hi @dani-mp ,

It looks like this is an ongoing issue on the API-side for batches on the Gemini API (Vertex API doesn't seem to be affected, though). There's more active discussions on the topic at https://discuss.ai.google.dev/search?q=batch%20order%3Alatest as well.

[dani-mp]

Hi @dani-mp ,

It looks like this is an ongoing issue on the API-side for batches on the Gemini API (Vertex API doesn't seem to be affected, though). There's more active discussions on the topic at https://discuss.ai.google.dev/search?q=batch%20order%3Alatest as well.

@speedstorm1 Should we switch to Vertex? Are there any tradeoffs we should be aware of?