10–20 seconds recommended. At least 5 seconds of clear continuous speech; stereo uses only the first channel. Singing and background audio are unsuitable.
Check the individual voice entry; not a blanket promise for every designed/system voice.
Data retention
Standard API inference prompts and responses are retained for up to 30 days, subject to legal/security exceptions; stateful files remain until deleted. CosyVoice is absent from the published list of ZDR-supported audio families.Source for Data retention (opens in a new tab)
wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference; run-task with model=cosyvoice-v3.5-flash. Same-region API key and compatible voice required.
Regions
China (Beijing). Deployment scope is not a country-only residency guarantee.
International-account USD tariff; do not mix with domestic CNY-account promotions.
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Speech synthesis
Speech synthesis1
$0.116
10,000 input characters
Rate conditions
Original China (Beijing) list rate, excluding promotions. Output has no separate charge. Chinese characters/kanji/hanja count as 2; other characters count as 1; SSML tags are excluded. Voice creation is a separate free workflow; inference is charged at this rate. No FX conversion.
Standard API inference prompts and responses are retained for up to 30 days, subject to legal/security exceptions; stateful files remain until deleted. CosyVoice is absent from the published list of ZDR-supported audio families.
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Inference from the distinct hosted release selector, official deployment documentation and reviewed Qwen/FunAudioLLM public-weight inventories. No corresponding weights or mapping to a published checkpoint were verified. This classification applies to the hosted tier, not to the public CosyVoice or Qwen model family. Private lineage and parameter counts remain undisclosed.
