
CosyVoice
situationalText-to-speech and voice-cloning model family from Alibaba Cloud.
Visit github.com →> **CosyVoice** — situational > Verified 2026-08-24 > Limitation: Token/character pricing can be confusing vs per-minute audio APIs. > > https://noemium.com/tools/cosyvoice/Add to kitOpen in NoemiumFact-check this catalog entry for CosyVoice (audio). Verdict: situational Verdict text: Best-in-class Chinese TTS naturalness, plus open-source weights you can self-host. Pricing is token-based and can be confusing compared with per-minute audio APIs; English quality is good but not market-leading. Limitations: - Token/character pricing can be confusing vs per-minute audio APIs. - Commercial rights and voice-cloning ethics need review. - English and non-Chinese quality are good but not market-leading. - Same Alibaba Cloud region/compliance friction as other Qwen services. Receipts: https://github.com/QwenAudio/CosyVoice https://www.alibabacloud.com/help/en/model-studio/voice-design-user-guide https://cloudprice.net/models/alibaba-cosyvoice-2 https://cloudprice.net/models/alibaba-cosyvoice-3-5-plus https://guijiagi.com/posts/ai-tts-2026-comparison/ Last verified: 2026-08-24 Find outdated prices, wrong claims, or missing limitations. Cite primary sources (docs, pricing pages, benchmarks, model cards). Do not use marketing pages.[](https://noemium.com/tools/cosyvoice/)
ledger entry · observed by @whysanesanders
Best-in-class Chinese TTS naturalness, plus open-source weights you can self-host. Pricing is token-based and can be confusing compared with per-minute audio APIs; English quality is good but not market-leading.
Strengths
- Top-tier naturalness for Chinese text-to-speech.
- Open-source weights that can be self-hosted.
- API and voice-creation options on Alibaba Cloud.
Known limitations
- Token/character pricing can be confusing vs per-minute audio APIs.
- Commercial rights and voice-cloning ethics need review.
- English and non-Chinese quality are good but not market-leading.
- Same Alibaba Cloud region/compliance friction as other Qwen services.
Use it when
- Chinese-language applications that need natural-sounding TTS.
- Teams that can self-host to keep voice data off shared cloud endpoints.
Skip it when
- You need simple per-minute pricing and predictable audio costs.
- English or non-Chinese output must be market-leading.
- Alibaba Cloud region and compliance requirements are a blocker.
Buy-check
- ·Price — paid, CosyVoice 2 from ~$0.287/10,000 characters; CosyVoice 3.5 Plus from ~$0.22/10,000 characters; CosyVoice voice creation free; Qwen-TTS voice creation $0.20/voice; free quota: 10 voice creations in first 90 days (Singapore region) as of 2026-08-24. Open the pricing page, not a screenshot.
- ·Free tier — Free tier exists — burn it first.
- ·Cancel fallout — What happens to your data/projects on cancel — check the ToS, not the FAQ.
- ·Seat math — Per-seat pricing multiplies quietly. Count seats before the annual plan.
- ·Who skips it — Also skipped when: You need simple per-minute pricing and predictable audio costs.; English or non-Chinese output must be market-leading..
Price trail — 2 revisionschangelog →
- 2026-W35CosyVoice 2 from ~$0.287/M input tokens; CosyVoice 3.…CosyVoice 2 from ~$0.287/10,000 characters; CosyVoice…
- 2026-W34CosyVoice 2 from ~$0.287/M input tokens; voice creati…CosyVoice 2 from ~$0.287/M input tokens; CosyVoice 3.…
Facts
- pricing
- paid1
- price note
- CosyVoice 2 from ~$0.287/10,000 characters; CosyVoice 3.5 Plus from ~$0.22/10,000 characters; CosyVoice voice creation free; Qwen-TTS voice creation $0.20/voice; free quota: 10 voice creations in first 90 days (Singapore region)
- free tier
- yes — paid to keep using
- open source
- yes
- api
- yes
- self-host
- yes
- category
- audio
- evidence
- source-verified
- Models
- locked
Data: trains on inputs unknown · local processing yes
Sources
- 1github.com/QwenAudio/CosyVoice[snapshot]
- 2alibabacloud.com/help/en/model-studio/voice-design-user-guide[snapshot]
- 3cloudprice.net/models/alibaba-cosyvoice-2[snapshot]
- 4cloudprice.net/models/alibaba-cosyvoice-3-5-plus[snapshot]
- 5guijiagi.com/posts/ai-tts-2026-comparison/[snapshot]
No source, no number.
Similar joball alternatives →
| tool | verdict | price | oss | self-host |
|---|---|---|---|---|
| CosyVoice | situational | paid | yes | yes |
| Deepgram | ship | freemium | no | no |
| ElevenLabs | ship | freemium | no | no |
| Suno | ship | freemium | no | no |
Facts from the catalog files — not scores you can buy. Quality and speed stay in the briefing, not in a fake 1–5 grid.
Deepgram
Speech-to-text API built for developers — fast, cheap, real-time.
APIFREE TIER
freemium$200 free credits; Nova-3 STT from $0.0048/min (promo)
ElevenLabsanchor
The reference standard for AI voice — TTS, cloning, dubbing and voice agents.
APIFREE TIER
freemiumfree tier (10k credits/mo, non-commercial); Starter $6/mo, Creator $22/mo
Suno
Full-song generation — vocals, lyrics, arrangement — from a text prompt.
FREE TIER
freemiumfree tier (10 songs/day, non-commercial); Pro $10/mo, Premier $30/mo