CosyVoice3
Half-billion-parameter multilingual TTS, RL-tuned for naturalness.
About
CosyVoice3 (Fun-CosyVoice3-0.5B) is FunAudioLLM's third-generation open TTS, adding reinforcement-learning optimization on top of the CosyVoice 2 architecture. It covers 9 languages plus 18 Chinese dialects with zero-shot voice cloning, 150ms bi-streaming latency, and instruction-based control over emotion, speed, and dialect.
The team's own eval set puts the RL variant at 0.81% Chinese CER — strong for a 0.5B model. As with most open TTS, the RL-tuned release is recent (December 2025) and third-party evaluations are still thin.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
CosyVoice 3.0 eval: 0.81% CER (ZH), 1.68% WER (EN) for RL variant — team-reported
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Official GitHub roadmap confirms Dec 2025 release; eval numbers from the team's own README; Exact release day not pinned — releasedAt left null; Manual review: strong 0.5B TTS; third-party evals still thin given recency. License: Open-source release; check the FunAudioLLM/CosyVoice repo for exact license terms.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Official GitHub roadmap confirms Dec 2025 release; eval numbers from the team's own README
- · Exact release day not pinned — releasedAt left null
- · Manual review: strong 0.5B TTS; third-party evals still thin given recency