Kokoro vs Qwen3-TTS
Both are Apache-2.0. Qwen3-TTS 1.7B is about 23x larger than Kokoro, and adds voice cloning and streaming with a vendor-claimed ~97 ms latency. It's a GPU-class model. We haven't benchmarked it yet.
Side by side
| Kokoro 82M | Qwen3-TTS 1.7B Base | |
|---|---|---|
| Organization | hexgrad | Alibaba Qwen |
| Parameters | 82M | 1.93B |
| License | apache-2.0 source | apache-2.0 source |
| Commercial use | Yes Apache-2.0; card welcomes production/commercial deployment | Yes Apache-2.0 |
| Stated memory | Not stated | Not stated |
| Runs on CPU | Yes | Unknown |
| Streaming | Unknown | Yes |
| Voice cloning | Unknown | Yes |
| Languages | 9 languages | 10 languages |
| Our fastest CPU result | RTF 0.1234 kokoro-onnx fp32 (onnxruntime CPU) voice=af_heart; Cloud x86 VM, 8 vCPU Intel Xeon, 16 GB RAM, no GPU | benchmark pending |
| Pi 5 | CPU supported, no device data | Unknown |
| CPU only | Measured by us | Unknown |
| Apple Silicon | Sourced statement | Estimate from parameter count |
| 8 GB GPU | Estimate from parameter count | Estimate from parameter count |
| 12–16 GB GPU | Estimate from parameter count | Estimate from parameter count |
| 24 GB+ GPU | Estimate from parameter count | Estimate from parameter count |
Each fact is sourced on the model's page. "Unknown" means the source doesn't say. Accuracy: Hugging Face Open TTS Leaderboard.
Key differences (sourced)
Which should you pick?
- Pick Qwen3-TTS if you have a GPU and need cloning or low-latency streaming in Chinese or English plus 8 more languages.
- Pick Kokoro if you're CPU-only or on a Raspberry Pi/N100, or need a tiny download.
Sponsor
Reach developers picking a speech model to self-host. Advertise here →
Sponsors never affect rankings, picker results or benchmark numbers.