ModelCards

Gemini 3.8 Live

Audio modelCard summarized

The default option for most low-latency voice agent experiences

Source: Gemini 3.8 Audio - Model Card

Use it via API

$0.005 / min
Code exampleCompare

Where to call it

  • Gemini API (Live API)gemini-3.8-live
  • Google Vertex AIgemini-3.8-live

Supports

  • Reasoning control
  • Web search
  • Image input
Full feature support

Good for

  • Real-time spoken conversation over continuous audio, video and text streams
  • Voice agents that run tool calls in the background while still talking
  • Grounding the conversation in live camera or screen input

Watch out for

Overview

Google · 04Live, front

3.8 Live

The default option for most low-latency voice agent experiences

Typical price
$0.005 / min
Audio input
Max length
15 min session
Languages
97 languages
Modes
speech-to-speech, real-time voice agent
Released
15 Sep 2026

One card covers Gemini 3.8 Live, Live Extended Thinking, Flash TTS and Flash-Lite TTS ("Gemini 3.8 Audio"), an addition to the Gemini 3 series based on Gemini 3 Pro. The Live models take continuous audio, video, image and text streams and reply with real-time speech. Architecture, training data and hardware are those of Gemini 3 Pro; the card itself reports no benchmark scores. Model information

gemini-3.8-live
Typical price
$0.005 / min
Max length
15 min session (audio only); 2 min with video
Languages
97 languages (launch post); the Live API docs list 99
Weights
Closed
Released
15 Sep 2026
  • speech-to-speech
  • real-time voice agent
  • background (async) function calling
  • live camera or screen input
  • input and output transcription
  • automatic language switching