Gemini Text to Speech: Models, Voices and Pricing
Gemini text to speech in October 2026: Gemini 3.8 Flash TTS and Flash-Lite TTS, languages, voice design, two-speaker limits and token pricing explained.
Gemini text to speech is the audio output of the Gemini API. The current models, gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts, became generally available in the Gemini API on 2026-09-22. Both take text and return audio, both support one or two speakers, and both are billed per token with a free tier. Prices change on 2027-01-01.
This guide maps the model names, prices, voices and limits as of October 2026, based on Google's developer pages and launch post fetched on 2026-10-17.
What Gemini text to speech does
Gemini text to speech turns text into speech. Google's speech generation guide describes it as suited to exact text recitation with fine-grained style control. The guide contrasts it with the Live API, which it describes as "designed for interactive, unstructured audio, and multimodal inputs and outputs."
The model accepts text only and returns audio only. The model page lists an input limit of 8,192 tokens and an output limit of 16,384 tokens for the Gemini API serving limit.
If you are building a voice agent that listens and answers in real time, the Live API is the relevant product. If you have a script, a set of lines or an article that must be read as written, text to speech is the better fit.
Gemini TTS models in October 2026
Google's deprecations page and changelog list these model IDs:
| Model ID | Status in the Gemini API | Languages (as stated by Google) | Output price per 1M audio tokens (Standard) |
|---|---|---|---|
| gemini-3.8-flash-tts | Generally available since 2026-09-22 | Over 130 | $9.00 to 2026-12-31, $18.00 from 2027-01-01 |
| gemini-3.8-flash-lite-tts | Generally available since 2026-09-22 | Over 100 | $6.00 to 2026-12-31, $12.00 from 2027-01-01 |
| gemini-3.1-flash-tts-preview | Preview; earliest shutdown 2026-11-17 | Not stated on the fetched page | $20.00 |
| gemini-2.5-pro-preview-tts | Preview; earliest shutdown 2026-11-17 | Not stated on the fetched page | $20.00 |
Google describes Flash as the "flagship model for creative speech." It describes Flash-Lite as built to replace gemini-3.1-flash-tts-preview for high-throughput production, such as dubbing, audio content and voice agents.
All four models support single-speaker and multi-speaker generation, according to the speech generation guide.
Gemini TTS pricing and cost per minute
Google's pricing page, last updated 2026-10-16 UTC, lists these Standard-tier rates in USD per 1M tokens:
- Gemini 3.8 Flash TTS: $0.50 input and $9.00 audio output through 2026-12-31. From 2027-01-01, $1.00 input and $18.00 audio output.
- Gemini 3.8 Flash-Lite TTS: $0.50 input and $6.00 audio output through 2026-12-31. From 2027-01-01, $1.00 input and $12.00 audio output.
The pricing page states that audio tokens correspond to 25 tokens per second of audio. That lets you estimate output cost per minute. This is a derived figure, and it covers output tokens only:
- Flash output at 1,500 tokens per minute is derived at $0.0135 per audio minute through 2026-12-31, and $0.027 from 2027-01-01.
- Flash-Lite output is derived at $0.009 per audio minute through 2026-12-31, and $0.018 from 2027-01-01.
Text input tokens cost extra. Other rate tiers from the same page:
- Batch and Flex rates are half the Standard rate.
- The Priority tier lists Flash TTS at $0.90 input and $16.20 output.
- Both 3.8 models have a free tier. Google says content used on the free tier is used to improve its products, and content on the paid tier is not.
Because the price is per token and not per character, a direct comparison with per-character vendors needs a conversion that Google does not publish. Treat the per-minute figures above as estimates.
Voices, languages and voice design
Google's speech generation guide lists 30 prebuilt voices, including Zephyr, Puck, Charon, Kore, Fenrir, Leda, Orus and Aoede. The guide also describes an Extended Voice Library, available through GET /v1beta/voices. Google's launch post uses a different count for its voice library, so this guide gives no total.
Language counts come from the speech generation guide: over 130 for Flash and over 100 for Flash-Lite, with a per-language table on the page. The launch post says "over 100 languages and dialects" for the family, so cite the guide figures and attribute them.
Google supports two voice-creation paths:
- Voice design: you describe a voice in natural language in AI Studio or through
POST /v1beta/voices. - Voice replication: you replicate a speaker's voice from reference audio and consent audio. Google's launch post says replication needs a 30-second sample and a verbal consent recording from the voice owner.
Google's limits for stored custom voices are 200 per project. A stored voice expires after one year without use, and stateless voice keys expire after 7 days. Google's launch post also says voice replication through AI Studio is not available in Illinois, Texas, the EEA, the UK, Switzerland and India.
How to direct delivery
Google gives you two ways to shape delivery:
- Turn-level style: put the overall style in
speech_metadata.style. Google's example is "whispered urgently." - Momentary events: place inline tags in the script for short pauses, laughs or sighs.
The text field is treated as a verbatim transcript. Do not put stage directions in it, because the model may read them aloud. Keep the style in speech_metadata.style and the events in the script.
Multi-speaker audio
Multi-speaker requests support up to two speakers, and they use prebuilt voices only. Single-speaker requests pass speech_config as an array. Multi-speaker requests pass a speakers list in the same structure. If your product needs three or more voices in one scene, you will need to split the audio into separate requests and join them yourself.
OndaLabs offers a two-speaker dialogue tool, which you can compare with Gemini's two-speaker limit. Read more about the dialogue feature.
Output formats and sample rates
The output format depends on how you call the API:
| Call type | Format | Details |
|---|---|---|
| Unary request | WAV (audio/wav) | 24 kHz, mono, 16-bit PCM with a RIFF header |
| Streaming | Raw PCM (audio/l16) | 24 kHz, no header |
| Other formats | audio/mulaw, audio/alaw | 8-bit G.711 |
A configurable sample_rate accepts values such as 24000, 16000 or 8000. The speech generation guide documents these options.
Gemini API or Google Cloud
Gemini TTS is available in two places, and the status differs. The Gemini API lists Flash and Flash-Lite as generally available. Google Cloud's pricing page lists the 3.8 models as Preview. Google's Cloud Gemini-TTS documentation says the 3.8 models are in Preview only through the Gemini Enterprise API and are not available through the Cloud Text-to-Speech API.
The Cloud documentation also lists general Gemini-TTS limits: text and prompt fields of at most 4,000 bytes each and at most 8,000 bytes combined, and output of about 655 seconds, with longer audio truncated. These are Cloud Gemini-TTS limits, not limits stated specifically for the 3.8 models. Check the Cloud page before you design around them.
Older models and retirement dates
Google's deprecations table lists an earliest shutdown date of 2026-11-17 for gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts and gemini-3.1-flash-tts-preview. The replacements named are gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts. Google says exact dates are communicated in advance.
The changelog is less clear. A 2026-09-18 entry says the 2.5 models "are not deprecated and will continue to be served until further notice," with access limited to previous users. The two Google pages disagree, so read both and confirm with Google before you rely on a date. The changelog and the deprecations table also give different release dates for gemini-3.1-flash-tts-preview, so this guide gives none.
Watermarking and safety
Google's launch post says every clip generated by the Gemini audio models is watermarked with SynthID, and replicated voices also receive C2PA credentials. These details come from the launch post, not from the API reference pages, so cite the blog if you need them in a policy document.
Where OndaLabs fits
Gemini TTS is an API model for developers, priced per token, with model names that change as previews move to general availability. If you are weighing it against OpenAI's per-character and per-token options, read our OpenAI text to speech guide first. OndaLabs is an AI voice studio for producing finished voice work.
The facts that matter for a creator or team choosing between the two:
- 30 prebuilt voices, usable in every supported language.
- Two model tiers: Onda Lite at 1 credit per character on all plans, and Onda Flash at 1.5 credits per character on paid plans.
- Custom voices from a recording you have permission to use, or designed from a text description.
- Plans run from Free ($0, 2,000 credits, about 2 minutes) to Business ($199 per month).
Google documents regional limits on voice replication in AI Studio. OndaLabs' published facts say custom voices come from a recording you have permission to use. Do not assume the two products share the same rules or regional availability. Check the pricing page and the text to speech page for current details.
Frequently asked questions
Which Gemini TTS model should I use, Flash or Flash-Lite?
Google positions Flash as the flagship model for creative speech, with direction and character work in mind. It positions Flash-Lite for high-volume, cost-efficient production such as dubbing and voice agents. Both are generally available in the Gemini API, and Flash-Lite is the lower-priced of the two on output tokens.
How much does Gemini TTS cost per minute of audio?
The pricing page does not give a per-minute price. Using its figure of 25 tokens per second of audio, output costs are derived at about $0.0135 per minute for Flash and about $0.009 per minute for Flash-Lite through 2026-12-31. Text input tokens are extra, and the prices double from 2027-01-01.
Is Gemini text to speech free?
Both 3.8 models have a free tier on the Gemini API. Google says content used on the free tier is used to improve its products, while paid-tier content is not. Check the pricing page for the current terms before you send sensitive material.
How many voices and languages does Gemini TTS support?
The speech generation guide lists 30 prebuilt voices and an Extended Voice Library. It states over 130 languages for Flash and over 100 for Flash-Lite. Google's launch post uses different voice and language wording, so cite the guide and check the per-language table for your language.
Can Gemini TTS produce multiple speakers or clone a voice?
Multi-speaker output supports up to two speakers, and those speakers must use prebuilt voices. Voice replication is also available, but it needs a 30-second sample and a verbal consent recording from the voice owner. Google's launch post lists regions where replication through AI Studio is not available.
Are the older Gemini 2.5 TTS models being retired?
Google's deprecations table gives an earliest shutdown date of 2026-11-17 for the 2.5 preview models. The changelog says those models are not deprecated and will continue to be served until further notice. Because the two pages disagree, confirm the status in the Gemini API documentation before you plan a migration.