← All models

CosyVoice-v3-Flash

Published Publisher Alibabacosyvoice-v3-flash

Overview

CosyVoice-v3-Flash is a Tongyi Lab speech synthesis and voice-cloning model. It converts text to audio, including through real-time streaming, and can create a custom voice from 5–20 seconds of reference audio. Alibaba reports improvements over earlier versions in naturalness, timbre, prosody, emotional expression, pronunciation accuracy and voice similarity. Supported languages include German, Spanish, French, Italian, Russian and Japanese. (help.aliyun.com)

Uses and limitations

It can be used for streaming text-to-speech and applications requiring a custom voice. The international model card lists text input and audio output; it does not list maximum input length, output length or context window. It lists function calling, structured output, web search, context caching and batch inference as unsupported. The card lists a limit of 180 requests per minute for both Beijing and Singapore. (alibabacloud.com)

Release

The supplied release date, November 17, 2025, matches the model’s entry in the official China-region release list. Alibaba’s release list records international deployment separately on February 9, 2026. Official release announcement. (help.aliyun.com)

API pricing

Prices below are for cosyvoice-v3-flash text-to-speech, checked November 17, 2026. Billing is based on input text characters; audio output is not separately charged. Regional and platform prices are distinct offerings, not currency conversions.

  • Alibaba Cloud China site, Beijing: CNY 1 per 10,000 input characters; no free quota is listed for this model. Singapore, international deployment: CNY 0.9541 per 10,000 input characters. China-site official pricing. (help.aliyun.com)
  • Alibaba Cloud international site, Singapore, international deployment: USD 0.13 per 10,000 input characters, with a listed free quota of 10,000 characters per account, valid for 90 days from the latest applicable activation, release or approval date. Beijing: USD 0.14335 per 10,000 input characters. International official pricing. (alibabacloud.com)

For CosyVoice, Alibaba’s China-site guide says creating a cloned voice is free; subsequent synthesis is billed under the applicable synthesis pricing. (help.aliyun.com)

Official model documentation

CosyVoice-v3-flash model card. (alibabacloud.com)