Gemini 3.1 flash live preview

Gemini API
1

Hi,

I’m currently using Gemini 3.1 Flash Live Preview for a real-time AI phone receptionist.

The model quality is good, but we’re still experiencing occasional instability, which makes it difficult to confidently deploy for production customer calls.

Is there any information available about:

  • When Gemini 3.1 Flash Live is expected to become production-ready / GA?
  • Whether the current Preview model is recommended for production voice workloads?
  • Whether Google is actively working on improving Live API reliability and consistency?
  • Whether a more stable successor to 3.1 Flash Live Preview is planned soon?
  • What would you recommend as the best temporary alternative for production telephony while 3.1 Flash Live is still in Preview?
  • If there is no stable Gemini Live model yet, is there a specific Gemini model, Google Cloud architecture, or combination of STT + LLM + TTS that you currently recommend for low-latency, customer-facing phone calls?

For real telephone calls, even an occasional missing response or unstable turn can break the conversation, so production stability is very important for our use case.

We would prefer to stay within the Gemini / Google ecosystem if possible, so any recommendation for a reliable interim production setup would be very helpful.

Thanks,
Maurits

2

I’m also interested. I noticed that at 7 a.m. Kazakhstan time, the model works very well! But in the evening, it works less well.

3

I feel your pain my friend. I too have a plugin that uses Gemin 3.1 Flash Live Preview, and it’s become unusable to my community. On most days, the AI doesn’t respond at all, and if it does, it comes out in cutoff delayed chunks. I can also confirm that it’s impacting multiple Gemini models, not just Flash Live 3.1 Preview. I fear even if/when it comes out of “Preview” to “GA”, that it will not change the performance issues. I’ve invested so much time into creating the infrastructure for my application and it pains me to have to find/implement other alternatives. I too would prefer to stay in the Gemini ecosystem. You’re not alone.

4

Thanks Charlie, really appreciate you sharing this. It’s reassuring to know I’m not the only one running into these issues.

It’s actually starting to slow down my customer acquisition as well, because I’m hesitant to put more customers onto something when I’m not fully confident it will behave reliably in production. Besides the delayed or missing responses, I’ve also seen things like the voice suddenly changing accent, which makes the experience feel unpredictable.

My biggest concern is exactly what you mentioned: even if 3.1 Flash Live moves from Preview to GA, there’s no guarantee these underlying reliability issues will disappear. I’ve invested a lot of time building around Gemini’s realtime stack, so switching isn’t something I want to do lightly either.

For now I’m looking at alternatives and fallback options, but I’d still prefer to stay with Gemini if Google can make it genuinely production-ready. If you find a setup that works reliably for you, I’d definitely be interested to hear what you end up using.