Batch API: every gemini-2.5-flash request fails with code 5 "Requested entity was not found"

Gemini API
1 
{"key":"81688..","error":{"code":5,"message":"Requested entity was not found."}}

with no response object at all. Same code, same request shape, same project ran totally clean on Sept 1 and Sept 2, a couple thousand requests in 150-request file batches, zero problems. I didn’t touch anything.

I spent a few hours last night trying to figure out what I broke before concluding I don’t think I broke anything. Sync generateContent on 2.5-flash with the same key works, 200 and a normal answer. models.list has models/gemini-2.5-flash in it. The uploaded jsonl files are accepted and still there with the right size, and since the keys come back in the error lines the file was obviously parsed. I tried file mode and inline mode (src as a list), both fail. Tried google-genai 2.6.0 and 2.23.0, both fail. Tried 2.5-flash-lite, fails the same way. Tried a second API key in a completely different Google project, also fails. Then I tried gemini-3.8-flash with the exact same 1-request batch on the same key and it went through fine, so the Batch API itself is clearly up, it just won’t run the 2.5 models.

The deprecations page shows 2.5-flash as GA with no shutdown date so I don’t think this is a retirement thing.

Batch IDs if it helps anyone look:

batches/3ssddpkftofkryvnwbn8rmu8r8l7fbhl0qto  (150 requests, created 01:20:04 UTC, 150 failed)
batches/wwwls26ec1y8b6l9pc4m5xvb1lnkz61y2uv0  (150 requests, 01:25:03 UTC, 150 failed)
batches/8jtsijpf9q4hlb3liy3px6510udou7re88wx  (1 request, file mode, SDK 2.6.0, 02:12 UTC, failed)
batches/hnyw7jyxs3jn9wd5xohi0cym6pod1g8k9l6u  (1 request, file mode, SDK 2.23.0, 02:12 UTC, failed)
batches/m9hg40i1otnh2eem88zadd39fx63ow08q3ks  (1 request, inline mode, 02:17 UTC, failed)
batches/66dqegyydhq1thtnh0p2zab6wuk07f4ib6cx  (1 request, different project, 02:17 UTC, failed)
batches/f057rz1e0qi0ji71s98uk45g2q8xzx492n5p  (1 request, gemini-3.8-flash, 02:19 UTC, worked)

This feels a lot like the July 9 thing where 2.5 flash started 404ing after a config change on your end and got rolled back within the hour. Is the batch backend maybe in the same state for 2.5 right now? I’ve got a few thousand requests sitting on hold until this clears up so I’d love to know if it’s being looked at or if I should just plan around it.

Happy to post the exact request body or more IDs if that’s useful.

2

Hey Mike, I’m also facing the same issue, and I’m wondering what we should do next. Do you have any idea how we can resolve this?

3

Whew I’m not alone. Not much we can do for now. Hope this is being read. 3.8 flash works for me bulk but I really prefer to have 2.5 flash working again.

4

Of course you are not alone, and in your place I would not hold my breath waiting for this to get fixed. I am convinced that this is not a bug, but a ‘feature’ designed to prod users to switch to more recent and more expensive models. You see, money talks…

The best remedy you can do is to implement a fallback switch in your code. For example, if your request for this model gets rejected 3 times after retries, then make the script to switch the model to a more recent one, and try again. That’s what I did in my Apps scrips, and it works more reliably this way.