Gemini 3.5 Transcribe Live

Model id: gemini/gemini-3.5-transcribe-live

Audio

Google’s live speech-to-text. You speak; it writes as you go.

Best for

live captions. Use the file model when you need speaker labels.

Provider: Google Gemini
Context window: Not published
Release date: 2026-08-26

Pricing

Input

$5.3981 / 1M

Output

$32.3885 / 1M

Amounts in Canadian dollars, converted at the Bank of Canada indicative rate (2026-09-21).

Performance

Reading

Last 7 days

Not enough data to display the chart.

Activity

Tokens

—

Top 5 tools · last 7 days

Tool activity appears only when at least 2 workspaces have used this model recently.

Availability

Availability (%)

Last 7 days

Operational probes: 99.98% over 7 days
Samples: 9,204

View service status

Quick start

Transcribe speech live with a tonia_ key and the exact model id below.

wscat

wscat -c "wss://pass.tonia.ca:8443/v1/realtime?provider=gemini&model=gemini-3.5-transcribe-live&mode=native&transcripts=1" \
  -H "Authorization: Bearer $TONIA_API_KEY"

Python — Live session

import os
import struct

from tonia import Tonia

pcm = struct.pack("<1600h", *([0] * 1600))
with Tonia(api_key=os.environ["TONIA_API_KEY"]) as client:
    with client.realtime.connect(
        provider="gemini",
        model="gemini-3.5-transcribe-live",
        mode="native",
        transcripts=True,
    ) as session:
        session.send_audio_append(pcm, mime="audio/pcm;rate=16000")
        session.send_audio_commit(mime="audio/pcm;rate=16000")
        print(session.recv(timeout=60.0))

TypeScript — Live session

import { Tonia } from "@tonia-router/sdk";

const pcm = new Uint8Array(3200);
const client = new Tonia({
  apiKey: process.env.TONIA_API_KEY,
});
const session = await client.realtime.connect({
  provider: "gemini",
  model: "gemini-3.5-transcribe-live",
  mode: "native",
  transcripts: true,
});
session.sendAudioAppend(pcm, "audio/pcm;rate=16000");
session.sendAudioCommit(undefined, "audio/pcm;rate=16000");
console.log(await session.recv(60_000));
session.close();

This model transcribes speech live. Send 16 kHz PCM audio. There is no spoken reply. Use the tonia SDK.

Create a key in the portal