What is Gemini 3.8 Flash TTS?
Gemini 3.8 Flash TTS turns a script into speech and reads it word for word. You pick one of 30 studio voices, search a library of more than 2,000, describe a new voice in plain words, or copy a real one from a short clip once the speaker records a consent statement. How each line sounds goes in a separate style field, and tags such as <sigh> or <short pause> place sounds at an exact word.
It is a current, stable model with no shutdown date, and with Flash-Lite TTS it replaces Gemini 3.1 Flash TTS, now a legacy preview. Google aims it at acting and dialects; Flash-Lite TTS costs less and suits voice agents and bulk audio.
Gemini 3.8 Flash TTS or Flash-Lite TTS?
They share one API and the same voices; Flash TTS acts better and speaks more languages, and Flash-Lite TTS costs less.
- Price
- $18 per million audio tokens against $12, or about $0.0045 against $0.003 per 10 seconds of speech.
- Languages
- More than 130 against 101; Finnish, Swedish, Thai and Swahili are among those only Flash TTS speaks.
- Best at
- Flash TTS: acting, dialects, two-voice scenes and long narration. Flash-Lite TTS: voice agents, read-aloud and bulk jobs.
- Switching
- Both take the same request format, so moving between them means changing the model name.
Who is Gemini 3.8 Flash TTS best for?
Gemini 3.8 Flash TTS is best for audiobooks and long narration, character voices for games and animation, two-voice podcasts, and scripts that need regional accents or less common languages.
Audiobooks and long narration
Google built it to hold the same voice, timbre and room tone across long narration, and names it the pick for audiobooks.
Character voices for games and animation
Voice design turns a description of age, accent and timbre into a saved voice you can reuse in every scene.
Two-voice podcasts and dialogue
One request can stage two speakers with natural turn-taking, plus short reactions such as |mhm| voiced over the other speaker.
Regional accents and less common languages
It covers more than 130 languages, including 29 that Flash-Lite TTS lacks, such as Finnish, Swedish, Thai and Swahili.
What's included with Gemini 3.8 Flash TTS?
Voice design, consented voice replication, a library of more than 2,000 voices, line-by-line style control with inline tags, two-speaker scenes, streaming, and SynthID watermarks.
- Voice design
- Describe a voice in plain words and get a reusable voice ID with a preview clip; a project can store 200 custom voices.
- Voice replication
- Copies a voice from a 10 to 30 second clip, plus the same speaker reading a consent statement in one of 30 supported locales.
- Voice library
- 30 studio voices plus more than 2,000 others, filterable by language, accent, gender, pitch and use.
- Line-by-line direction
- The text is read verbatim; delivery goes in a per-line style field, and inline tags add laughs, sighs, pauses and IPA pronunciations.
- Two-speaker scenes
- Up to two prebuilt voices in one request, with backchannels and overlapping speech written inside pipes.
- Streaming and phone formats
- Streams raw 24 kHz audio while it is generated, and can return WAV, mu-law or A-law at rates down to 8 kHz.
- SynthID watermark
- Google marks every clip its Gemini audio models generate with an inaudible SynthID watermark.
How does Gemini 3.8 Flash TTS work?
Choose or build a voice
Pick a studio or library voice, design one from a description, or replicate one with consent in Google AI Studio or through the API.
Write the script as it is spoken
Put only the words to be spoken in the text; Google suggests trying it with an empty style first.
Direct the lines
Add a short style to the lines that need it, and inline tags for sighs, laughs or pauses.
Generate or stream
Get a WAV file back from each request, or stream raw audio to play it while it is being made.
How much does Gemini 3.8 Flash TTS cost?
Gemini 3.8 Flash TTS is free on the Gemini API's rate-limited free tier; paid use costs $1 per million text tokens in and $18 per million audio tokens out, about $0.0045 per 10 seconds of speech, or half that on Batch or Flex.
- Free, rate-limited text input and audio output
- Your content may be used to improve Google's products
- No Batch or Flex jobs
- Text input, audio output
- Half the Standard rate, about $0.00225 per 10 seconds of speech
- Cached input $0.125 on Batch, $0.05 on Flex
- Text input, audio output; about $0.0045 per 10 seconds of speech
- Cached input $0.25, plus $1.00 per million tokens per hour of storage
- Paid-tier content is not used to improve Google's products
- Text input, audio output
- About $0.0081 per 10 seconds of speech
- Cached input $0.45
Prices are for the Gemini Developer API in US dollars. Audio counts as 25 tokens per second, so Standard works out to about $1.62 per hour of speech. Paid use needs a linked billing account and at least $5 of prepaid credit, which expires after 12 months. Google AI Studio is free to use, and prices on Gemini Enterprise Agent Platform can differ. Full pricing on ai.google.dev. We checked these prices on .
How do readers rate Gemini 3.8 Flash TTS?
No one has rated Gemini 3.8 Flash TTS yet. Be the first.
Sign in to rate. One rating per account.
Gemini 3.8 Flash TTS pros and cons
It is expressive, easy to direct and about a third of Eleven v3's price per minute, but early users say it can overact, miss the prompt or drift between voices.
Pros
- Far cheaper than ElevenLabs for long audio, a point Hacker News users picked out: Standard works out to about $0.027 a minute of speech, against about $0.08 a minute for Eleven v3.
- Early users on Hacker News and Reddit call the delivery very expressive; one developer who swapped it in for Eleven v3 rated it on par.
- Easy to direct: users say the per-line style field and inline tags make the tone easier to get than with Gemini 3.1 Flash TTS or GPT-Live.
- Google reports first place on Hume AI's Voice Design Benchmark (71.4) and the lead in accent modelling (60.8).
Cons
- Output does not always match the direction: Hacker News users found Google's own demos missing parts of the voice prompt and skipping tags such as <chuckles>, and one <sigh> came out sounding like a moan.
- Some listeners find the delivery overacted, with exaggerated emotion where they wanted a plain, even voice.
- Voice consistency is mixed: some Reddit users say a long text split into parts comes back in slightly different voices, while others found it steadier than 3.1 Flash TTS.
- One request holds at most two speakers, both prebuilt voices, so scenes with designed or cloned voices are built turn by turn; voice replication in Google AI Studio is not offered in the EEA, UK, Switzerland, India, Illinois or Texas.
What are the best Gemini 3.8 Flash TTS alternatives?
Three tools on Recaply do Gemini 3.8 Flash TTS's job. Qwen3-TTS CustomVoice 1.7B is open source, so you can run it on your own server.
Eleven v3FreeChoose it when you want ElevenLabs' app, cloning plans and 70+ languages in one place; it costs $0.08 per 1,000 characters on the API, about three times Flash TTS per minute, and caps each request at 5,000 characters.
Gemini 3.8 Flash-Lite TTSFreeChoose it when you build voice agents, read-aloud features or bulk audio; it uses the same API and voices at $12 instead of $18 per million audio tokens, but covers 101 languages instead of 130.
Qwen3-TTS CustomVoice 1.7BOpen SourceChoose it when you would rather run speech on your own GPU with no per-minute bill; the weights are free under Apache 2.0, but it offers 9 preset voices in 10 languages.
Questions about Gemini 3.8 Flash TTS
What is Gemini 3.8 Flash TTS?
Gemini 3.8 Flash TTS is Google's top text-to-speech model on the Gemini API, built for audiobooks, games and podcasts, with voices you can design, clone with consent and direct line by line.
Who is Gemini 3.8 Flash TTS for?
Gemini 3.8 Flash TTS is best for audiobooks and long narration, character voices for games and animation, two-voice podcasts, and scripts that need regional accents or less common languages.
Is Gemini 3.8 Flash TTS free?
Yes. Gemini 3.8 Flash TTS has a free Free tier plan: free, rate-limited text input and audio output.
How much does an hour of audio cost with Gemini 3.8 Flash TTS?
About $1.62 at Standard rates. Google prices audio output at $18 per million tokens, which it puts at $0.0045 per 10 seconds, and Batch or Flex halves that. The text you send costs $1 per million tokens, small next to the audio.
Is there an open source alternative to Gemini 3.8 Flash TTS?
Yes. Qwen3-TTS CustomVoice 1.7B is an open source tool that does the same job, so you can run your own.
What are the best Gemini 3.8 Flash TTS alternatives?
The closest alternatives on Recaply are Eleven v3, Gemini 3.8 Flash-Lite TTS and Qwen3-TTS CustomVoice 1.7B.
Where can I try Gemini 3.8 Flash TTS?
On its official website, ai.google.dev.
Can Gemini 3.8 Flash TTS clone my voice?
Yes, with consent. Voice replication needs a 10 to 30 second clip of the speaker and a recording of the same person reading Google's consent statement in one of 30 supported locales. In Google AI Studio it is not offered in the EEA, UK, Switzerland, India, Illinois or Texas.
How long can one Gemini 3.8 Flash TTS request be?
Up to 8,192 tokens of text in and 16,384 tokens of audio out on the Gemini API. Audio counts as 25 tokens a second, so one request gives about 11 minutes of speech, and longer books are generated in parts.
How do I move from Gemini 3.1 Flash TTS to Gemini 3.8 Flash TTS?
Change the model name, then move stage directions and speaker labels out of the text into speech_metadata, since 3.8 reads the text word for word. Keep only momentary sounds such as <laugh> inline, and expect unary responses as WAV rather than raw PCM.
Does Google use my text to improve Gemini 3.8 Flash TTS?
On the free tier, Google says your content is used to improve its products. On the paid tier it is not, so use a paid key for scripts that are private or under contract.
Our verdict
Gemini 3.8 Flash TTS is the Gemini voice model to choose when the performance matters more than the lowest price.
It pairs line-by-line direction with voice design, consented cloning and more than 130 languages, and it costs about a third of Eleven v3 per minute of speech.
- Skip it for
- Real-time voice agents and bulk read-aloud work, where Flash-Lite TTS is cheaper and built for speed.
- Check first
- Run your own scripts in Google AI Studio first, since early users say it can overact or wander from the voice prompt.
Gemini 3.1 Flash TTS
Gemini 3.1 Flash Live
Eleven v4
MAI-Voice-2
MisoTTS
Qwen3-TTS