Summary
The gemini-3.1-flash-live-preview model is currently available only
on AI Studio (generativelanguage.googleapis.com — global routing).
There is no Vertex AI version, regardless of region.
For production use we need it on Vertex AI in an EU region
(europe-west4 / europe-west1 / europe-north1) — GDPR requires that
inference for EU end-users stays within the EU. Vertex AI is the only
Gemini surface that offers regional endpoints with a data residency
guarantee; AI Studio’s global routing does not satisfy this.
Question
Is gemini-3.1-flash-live-preview (or any 3.1 Live variant — text-
or native-audio-based) on the Vertex AI roadmap? If yes, what is the
expected timeline and which EU regions will be supported?
The forum thread “Gemini 3.1 Flash Live Preview on Vertex AI, when?”
1 has been open with no official response. Adding context here in
case it helps escalate visibility.
Also waiting for 3.1 Flash Live on Vertex AI EU. We tried
gemini-2.5-flash-native-audio` as a fallback but ran into the
control-token leakage bug (model emits `` sequences instead
of audio).
Looks like we’ll have to fall back to Azure + gpt-realtime. Word
recognition is noticeably worse than Gemini, but it’s stable, doesn’t fall into degenerate states mid-call, and ships with proper EU region routing for GDPR.
Really hoping Google ships 3.1 Live on Vertex AI EU soon - we’d
much rather stay on Gemini for the audio quality.
Hey,
I am also using Live for my application but I came to a different conclusion about GDPR.
If I may ask why would not having an EU hosted model be GDPR breaking?
Isn’t it possible to satisfy GDPR with US hosted models?
My research pointed out that this is only a requiremenent when sensitive data is involved and then it is up to debate if you are actually handling e.g. medical data or just relaying information.