Guides · both · chinese text to speech api

Chinese text-to-speech prices, on one unit, against ElevenLabs

Published 16 Oct 2026. Updated 16 Oct 2026. Researched 16 Oct 2026.

Voice cloning is where a “cheap Chinese API” post usually mixes three units: per character, per million characters, and per UTF-8 byte. This page only charts rates we could put on “US dollars per 1,000 characters” without pretending a byte is a character.

The chart

MiniMax’s docs list speech-2.8-turbo at $60 per million characters and speech-2.8-hd at $100 per million, for both sync and async. ElevenLabs’s API page lists v4 Turbo at $0.04 per 1,000 characters and v4 at $0.08, and it says the per-character rate is the same across plans. QwenCloud lists several TTS ids per 10,000 characters, including cosyvoice-v3-flash at $0.13 and cosyvoice-v3-plus at $0.26. The chart divides those published figures onto one unit. The extra column is the figure as printed.

We did not store an Artificial Analysis speech arena. There is no Elo chart. A lower bar here is a lower list price, not a preferred voice.

Fish Audio is a different meter

Fish’s own docs price s2.1-pro at $15 per million UTF-8 bytes, and s2.1-pro-free at $0. They say 1 million UTF-8 bytes is about 180,000 English words. Chinese text is not one byte per character, so we will not convert Fish onto the chart. Voice design is $0.01 per successful request. ASR is $0.36 per audio hour. The licence for the open CosyVoice code is Apache 2.0. The hosted cosyvoice-v3 ids on QwenCloud are a bill, not that GitHub repo.

MiniMax also charges $1.50 when a rapid clone is first used for synthesis, and $3 when a designed voice is first used. Those are not in the per-character bar. Audio subscriptions start at $5 a month for 100,000 audio points. We are not converting points into characters. The docs do not give us that ratio.

ByteDance speech, without a made-up rate

BytePlus is the international door. The free-trial page lists SeedTTS text-to-speech at 20,000 characters, voice replication at 20,000 characters plus one voice, and SeedASR at 20 hours, for eligible new users, and it says the quota does not expire but can change. The ModelArk pay-as-you-go table we captured on the product page does not include a Seed-TTS price. An older BytePlus Voice “Audio 1.0” page prices audio at $0.15 per minute and was last updated 14 Jul 2026. That product name is not Seed Speech 2.0, so it is not used as the Seed rate. The BytePlus guide is where the console and the card rules live.

Privacy

A cloned voice is biometric data. MiniMax, Alibaba, Fish, and ByteDance are not the same company, and none of them is a reason to upload someone else’s recording. ElevenLabs is the Western comparator on price only. If you can run CosyVoice locally, the audio never hits those APIs. We did not measure a real-time latency number, so none is printed. ElevenLabs’s page claims about 100 ms for v4 Turbo. That is their claim, not our stopwatch.

Related: BytePlus, music, MiniMax M3 for the text model, which is not this speech price.