For Developers/Models/Gemini 3.8 Live
Closed SourceGoogle DeepMindReleased 2026-09

Gemini 3.8 Live

Most advanced live dialogue model for natural, real-time voice conversations

Context window

128K

Input / 1M tokens

$3.00

Output / 1M tokens

$12.00

Provider

Google DeepMind

Data verified 2026-09-16

Gemini 3.8 Live is a natively multimodal audio model built for cost-efficient, high-volume voice interactions. It processes visual inputs in near real time, automatically detects and transitions between 97 supported languages mid-conversation, and executes tool calls in the background without interrupting dialogue. Based on Gemini 3 Pro, it delivers fluid conversational intelligence with visual grounding.

Capability index

Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.

Coding
0
Reasoning
3
Math
2
Multimodal
8
Long context
6
Speed
9
Cost efficiency
8

How to access it

Available via Gemini API, Google AI Studio, Search Live, Gemini Live, and Gemini Enterprise. Voice-native model for real-time conversational AI.

Strengths

  • Near real-time audio processing with low latency
  • Supports 97 languages with automatic mid-conversation detection
  • Visual input processing in real time
  • Background tool execution without dialogue interruption
  • Cost-efficient for high-volume voice applications

Best for developers who...

Production voice agentsReal-time conversational AICustomer service automationMulti-language voice applications

When to choose it (and when not to)

Reach for Gemini 3.8 Live when...

  • When building cost-efficient voice agents
  • For real-time latency-sensitive applications
  • When visual grounding is needed during conversation

Look elsewhere if...

  • High-complexity reasoning tasks (use 3.8 Live Extended Thinking instead)
  • Tasks requiring deep multi-step reasoning

How to use it

  • Speak naturally; the model is optimized for fluid conversation
  • Leverage background tool execution to keep dialogue uninterrupted
  • Use visual context for enhanced contextual understanding

Quickstart

Python
from google import genai
client = genai.Client()
response = client.models.generate_content(
    model='gemini-3.8-live',
    contents='Hello, how can I help you today?'
)

Requires Google API credentials and Gemini API access

API model id: gemini-3.8-live

Benchmarks

BenchmarkScoreNotes
Speech to Speech Quality Index (Artificial Analysis)80.1%Ranks among top voice quality models
Big Bench Audio97.7%Strong audio processing performance

Source: Artificial Analysis

Compare Gemini 3.8 Live

Compare Gemini 3.8 Live with any other model

Build a comparison →
All model comparisons →

Learn the concepts