whatmodel.sh / tts / CosyVoice v3 Plus
CosyVoice v3 Plus
↗ share
+ contribute
CosyVoice v3 Plus is a closed-weights text-to-speech model from Alibaba Cloud, released September 2025.
Catalogue identity notes
Source product model cosyvoice-v3-plus; voice configured default Chinese voice. June 2026 invocation. Do not map Azure Neural to Azure HD or Polly Neural to Polly Generative.
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 CosyVoice v3 Plus, out of 148 with a score.
| # | model | score | where |
|---|---|---|---|
| 53 |  MaskGCT | 59±1 | local |
| 54 |  MGM-Omni-TTS 2B (0927) | 59±4 | comparison only |
| 55 | CosyVoice v3 Plus | 58±4 | api $26.00/M chars |
| 56 |  Octave TTS (Octave 1) | 58±4 | api $50.00/M chars |
| 57 |  Sonic 3 | 58±2 | api $65.00/M chars |
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
other tests1 not in a task · 94% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CN-NewsTTS v0.1 · strict pronunciation accuracy | 47.2% ✎ | █··▄·█··▄··▄ beats 2/6 | 94% |
$ where to run it
Through an API
api · Alibaba Cloud Model Studio · $26.00/M chars · www.alibabacloud.com ↗ · price sourceUSD per 1M characters: Model Studio lists cosyvoice-v3-plus at $0.26 per 10,000 input characters, International (Singapore) deployment; output free (x100). Each Chinese character counts as 2 characters. Product: Alibaba Cloud Model Studio speech synthesis. API model: cosyvoice-v3-plus. Settings: International deployment. Price basis: published list rate. Exclusions: free quota; China (Beijing) lists $0.286706 per 10,000. Currency: USD. Alibaba Cloud official pricing page, observed 2026-09-25.
$ how the score was worked out · CN-NewsTTS Bench v0.1 94%
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 |
|---|---|---|
 CN-NewsTTS Bench v0.1 | 1 | 94% |
| prior · pull toward the middle | 6% | |
| weighted average | = 58.0 |
$ 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