Last updated: October 13, 2026

Gemini 3.8 Flash TTS: pricing, voices and API guide

Google released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on September 23, 2026. Flash TTS is the expressive model for narration, audiobooks and character voices; Flash-Lite TTS is the cheaper one for dubbing and voice agents. This page covers price per hour, voices, code and the limits to know before you build on it.

Gemini 3.8 Flash TTS at a glance

Source: Gemini 3.8 Flash TTS model page

Source: Google TTS launch post

Gemini 3.8 Flash TTS pricing

Audio output is billed at 25 tokens per second, so price per hour is easy to work out. Launch prices run through December 31, 2026; standard prices start January 1, 2027.

Per-hour figures are our arithmetic from Google’s per-token prices and exclude the small text input cost. The free tier covers both models; free-tier inputs may be used to improve Google products.

Source: Gemini API pricing

Python example

1. Create a key in Google AI Studio and save it as GEMINI_API_KEY.
2. pip install -U google-genai.
3. Put the words to speak in text and the delivery (emotion, pace, tone) in the style annotation, not in the text.
4. Pick a voice and write the base64 audio to a .wav file.

# pip install -U google-genai
import base64
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-3.8-flash-tts",
    input=[{
        "type": "user_input",
        "content": [{
            "type": "text",
            "text": "Have a wonderful day! <short pause> See you tomorrow.",
            "annotations": [{"type": "speech_metadata", "style": "cheerful and friendly"}],
        }],
    }],
    response_format={"type": "audio"},
    generation_config={"speech_config": [{"voice": "Kore"}]},
)

with open("out.wav", "wb") as f:
    f.write(base64.b64decode(interaction.output_audio.data))

Source: Speech generation docs

curl example

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.8-flash-tts",
    "input": [{
      "type": "user_input",
      "content": [{
        "type": "text",
        "text": "Have a wonderful day!",
        "annotations": [{ "type": "speech_metadata", "style": "cheerful and friendly" }]
      }]
    }],
    "response_format": { "type": "audio" },
    "generation_config": { "speech_config": [{ "voice": "Kore" }] }
  }' | jq -r '[.steps[] | select(.type=="model_output") | .content[] | select(.type=="audio")] | last | .data' \
     | base64 --decode > out.wav

Source: Speech generation docs

Two-speaker dialogue and inline cues

Name each speaker once in speech_config, then tag every turn with its speaker and an optional style. Short events go inline in the text: <laughs>, <sigh>, <gasp>, <short pause>, and backchannel sounds such as |mhm| or |yeah|.

# Two speakers in one request (the most a request allows with prebuilt voices)
generation_config={"speech_config": {"speakers": [
    {"speaker": "Host", "voice": "Kore"},
    {"speaker": "Guest", "voice": "Puck"},
]}}

input=[{"type": "user_input", "content": [
    {"type": "text", "text": "Welcome back to the show. <laughs> Big week.",
     "annotations": [{"type": "speech_metadata", "speaker": "Host", "style": "warm, upbeat"}]},
    {"type": "text", "text": "|mhm| It really was.",
     "annotations": [{"type": "speech_metadata", "speaker": "Guest", "style": "relaxed"}]},
]}]

Text is read as a verbatim script, so stage directions written into the text itself may be spoken aloud. Keep them in style.

Source: Speech generation docs

Source: Google TTS launch post

Voices, voice design and voice cloning

• 30 prebuilt voices such as Kore, Puck, Zephyr and Charon, plus an extended library of 2,000+ production voices with regional accents (Mexican Spanish, Quebec French, Scots English and more).
• Voice design: describe a voice in plain language (role, accent, character) and save it for reuse.
• Voice replication: clone a voice from about a 30-second sample, with consent checks, SynthID watermarking and C2PA credentials.
• In AI Studio, voice replication is not available in Illinois, Texas, the EEA, the UK, Switzerland or India.

Source: Google TTS launch post

Source: Gemini 3.8 Flash TTS model page

How good is it?

Google cites Hume AI’s Voice Design Benchmark, where Flash TTS ranks first with 71.4 overall and 60.8 on accent modelling, and first and second place (Flash, then Flash-Lite) on Hume’s overall quality index. In Voice Arena blind tests, both models rank near the top in Japanese, Brazilian Portuguese, Vietnamese, Arabic, Mexican Spanish and Hindi. These are vendor-cited results; listen to your own script before choosing.

Source: Google TTS launch post

Flash TTS or Flash-Lite TTS?

• Flash TTS. Audiobooks, podcasts, ads and game characters, where delivery matters more than price.
• Flash-Lite TTS. Voice agents, dubbing and high-volume notifications, about a third cheaper per hour.
• Neither. Live, two-way voice conversations: use Google’s Live API models instead; the TTS models do not support it.

Gemini 3.8 Flash API guide · All model prices

Source: Google TTS launch post

Source: Gemini 3.8 Flash TTS model page

Independent guide, not affiliated with Google or Anthropic. Prices and benchmarks change; check the linked official pages before you commit to a budget.