whatmodel.sh / tts / Gemini 3.1 Flash TTS Preview
Gemini 3.1 Flash TTS Preview
↗ share
+ contribute
Gemini 3.1 Flash TTS Preview is a closed-weights text-to-speech model from Google, released April 2026.
Catalogue identity notes
Exact model gemini-3.1-flash-tts-preview; voice Kore. Dataset label Gemini 3.1 Flash TTS. No API immutable build/weight SHA supplied.
Model information sources
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of Gemini 3.1 Flash TTS Preview, out of 148 with a score, then more from Google.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 6 | Gemini 2.5 Pro Preview TTS | 69±3 | comparison only | ||
| 7 |  Qwen-Audio-3.0-TTS-Plus | 69±2 | api $20.00/M chars | ||
| 8 | Gemini 3.1 Flash TTS Preview | 69±2 | api $50.00/M chars | ||
| 9 |  dots.tts MF 2B (MeanFlow, NFE 4) | 68±3 | local | ||
| 10 |  Raon-OpenTTS-1B | 68±2 | comparison only | ||
| more from Google | |||||
| 3 | Gemini 2.5 Flash Preview TTS | 72±3 | comparison only | ||
| 22 |  Chirp 3: HD | 65±4 | api $30.00/M chars | ||
| 116 | Gemini 3.8 Flash TTS | 37±3 | api price unknown | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning▁·▁▁▁▁▁▁▇█▃▃ #23 of 37 * · 5.2% of its score · 1 test
Trails E2 TTS Base (F5-TTS reproduction), DMOSpeech 2, Qwen3-TTS 12Hz 0.6B Base here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| nsanku TTS Benchmark · 7 Ghanaian languages mean CER (GhanaNLP) | 20.0% ✎ | █··········▄ beats 0/2 | 5.2% |
other tests13 not in a task · 93% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ZTTS1-Eval in-the-wild · 10-language mean WER | 2.6% ✎ | ▄·····█··▄██ beats 5/7 | 15% |
| ZTTS1-Eval Clean (FLEURS-R) · 9-language mean WER | 4.0% ✎ | ▄·····▄█▄█·▄ beats 4/7 | 13% |
| ZTTS1-Eval in-the-wild · 10-language mean MSR-UTMOS | 3.51 / 5 ✎ | █··▄█▄····▄▄ beats 7/7 | 13% |
| ZTTS1-Eval Clean (FLEURS-R) · 9-language mean MSR-UTMOS | 3.55 / 5 ✎ | ▃▃█···▃···▃▃ beats 6/7 | 11% |
| Wayu Thai TTS eval · 1,531-item Challenge-Set keyword accuracy | 79.8% ✎ | █···█······█ beats 2/2 | 7.4% |
| Wayu Thai TTS eval · English CER on 500 held-out LibriTTS lines (Whisper Large V3) | 0.8% ✎ | █······█···█ beats 2/2 | 7.2% |
| RW-Voice-EQ TTS · Acting / Role-Fit (human) | 3.97 / 5 ✎ | ▄·····█····▄ beats 3/3 | 6.8% |
| Wayu Thai TTS eval · 500-utterance Thai CER (Typhoon Whisper Large V3) | 3.3% ✎ | █······█···█ beats 2/2 | 6.6% |
5 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ZTTS1-Eval · FLEURS-R English TTSDS2 prosody | 87.6% ✎ | █··███·····█ beats 3/4 | 4.4% |
| RW-Voice-EQ TTS · Reliability (human) | 3.82 / 5 ✎ | █·▄······▄·▄ beats 3/4 | 4.3% |
| RW-Voice-EQ TTS · Expressiveness (human) | 3.64 / 5 ✎ | ███···█····█ beats 4/4 | 2.8% |
| Arabic TTS Arena · community blind A/B rating, 2026-09-24 snapshot | 1147.5 ✎ | ▂▂··▃·█▅···▃ beats 12/12 | 2.3% |
| Amharic TTS · 87-prompt CER (Chirp 3 judge) | 2.9% ✎ | ▄·····▄·▄·██ beats 6/6 | 0.0% |
$ where to run it
Through an API
api · fal.ai · $50.00/M chars · fal.ai ↗ · price sourceUSD per 1M input characters: fal prices its Gemini 3.1 Flash TTS endpoint (fal-ai/gemini-3.1-flash-tts; input schema: 'Input for Gemini 3.1 Flash TTS preview speech generation') at $0.05 per 1000 characters (x1000; endpoint billing record: billing_unit 1000 characters, partner API). Google's own Gemini API prices this model per token only, which is not converted. Product: fal text-to-speech endpoint (Google partner API). API model: fal-ai/gemini-3.1-flash-tts (Gemini 3.1 Flash TTS preview). Settings: standard. Price basis: published list rate, per-1K to per-1M scaling only. Currency: USD. fal.ai model page, observed 2026-09-25.
$ how the score was worked out · ZONOS2 Technical Report (ZTTS… 51%, Building and Evaluating Fixed… 21%, RW-Voice-EQ Bench (Hume AI) 14% and 4 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 ZONOS2 Technical Report (ZTTS1-Eval results) | 4 | 51% |
| Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech (Wayu Research / Paxa Labs / Typhoon; arXiv:2609.03502v1) | 3 | 21% |
| RW-Voice-EQ Bench (Hume AI) | 3 | 14% |
 nsanku TTS Benchmark (GhanaNLP community Space) | 1 | 5% |
| ZTTS1-Eval results snapshot (Zyphra) | 1 | 4% |
| Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24 | 1 | 2% |
| Addis AI Amharic TTS benchmark | 1 | 0% |
| prior · pull toward the middle | 1% | |
| weighted average | = 68.5 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json