The Chinese TTS Stack, Licensed: CosyVoice (Apache) vs ChatTTS (Non-Commercial) — What You Can Actually Ship
ElevenLabs is great — but the best open TTS in 2026 is Chinese. Here's which Chinese TTS you can legally ship, with verified licenses.
TL;DR — Everyone uses ElevenLabs (from ~$22/mo). The open-source alternative that actually rivals it is CosyVoice 3 by Alibaba's Qwen team — Apache-2.0, free, runs locally. But the second-most-hyped Chinese TTS, ChatTTS, is CC BY-NC 4.0 — non-commercial: you cannot ship it. Verified via the GitHub API on 2026-09-03.
The data
| ElevenLabs | CosyVoice 3 (QwenAudio) | ChatTTS | |
|---|---|---|---|
| License | Proprietary | Apache-2.0 ✅ | CC BY-NC 4.0 🔴 |
| Price | ~$22–330/mo | Free (self-host) | Free (self-host) |
| Runs locally | No | Yes (MLX port) | Yes |
| Stars | — | 23,400+ | 33,000+ |
| Verified | — | 2026-09-03 (GitHub API) | 2026-09-03 (GitHub API) |
The "because Z"
- License is the product. ChatTTS has more stars and sounds great — but its model weights are CC BY-NC. One DMCA later, your voice feature is gone. CosyVoice is Apache-2.0 end to end: ship it in a commercial app today.
- It's the Alibaba/Qwen team. Same org that maintains the Qwen models Western devs already run. Active releases, an MLX port for Apple Silicon, and CosyVoice 3 supports voice design + controllable cloning.
- Cost curve. For volume use (10+ voices, audiobooks, dubbing pipelines), self-hosted CosyVoice on a $699 Mac mini amortizes to near zero vs ElevenLabs monthly fees.
The trap to avoid
Check the license field, not the star count. Several hyped Chinese TTS projects (ChatTTS, GPT-SoVITS ecosystem variants) carry non-commercial terms. Chinese licenses are fluid — MinerU went AGPL→Apache-2.0 in 2026, and some projects flip the other way. We re-verify every license via the GitHub API on every update; never assume.
Verdict
- You need a commercial voice pipeline → CosyVoice 3 (Apache-2.0, local, free at scale).
- You prototype with zero risk and want ElevenLabs polish → keep ElevenLabs — it's still the fastest path, and their API is excellent.
- Never build production on a CC BY-NC model, however good the demos sound.
Data verified 2026-09-03 · Sources: GitHub API (openbmb/ChatTTS, QwenAudio/CosyVoice), project pages.
Was this article helpful?
Let us know to improve our AI generation.