Live API Real-Time Streaming — Code Execution & Live API

Bidirectional Real-Time AI

The Gemini Live API — powered in October 2026 by gemini-3.8-live (plus gemini-3.8-live-extended-thinking, 131,072 input / 65,536 output tokens, 16 kHz PCM audio in / 24 kHz audio out, and optional Live Avatar synchronization) — enables real-time, bidirectional streaming of audio, video, and text over a persistent WebSocket connection (via client.aio.live.connect(model="gemini-3.8-live", ...) in google-genai or direct WebSockets).

Key Capabilities

WebSocket Connection

import asyncio
import websockets
import json

async def live_session():
    uri = "wss://generativelanguage.googleapis.com/ws/google.ai.generativelanguage.v1beta.GenerativeService.BidiGenerateContent"
    async with websockets.connect(uri) as ws:
        # Send setup message
        await ws.send(json.dumps({
            "setup": {
                "model": "models/gemini-3.8-live",
                "generation_config": {
                    "response_modalities": ["AUDIO"],
                    "speech_config": {"voice_config": {"prebuilt_voice_config": {"voice_name": "Puck"}}}
                }
            }
        }))
        # Stream audio chunks bidirectionally
        await ws.send(json.dumps({
            "realtime_input": {"media_chunks": [{"data": audio_b64, "mime_type": "audio/pcm"}]}
        }))
        # Receive streaming response
        response = await ws.recv()
        print(json.loads(response))

asyncio.run(live_session())

Barge-In Support

Barge-in allows users to interrupt the model while it is still generating a response. The model detects the interruption, stops its current output, and immediately processes the new input - creating a natural, conversational experience.