best model by language
Benchmarked on the FLEURS test split at Q8_0. Error is CER — lower is better. Speed is × realtime on Handy target hardware.
accuracy vs speed
Mandarin Chinese · 20 models
AMD Ryzen 4750U GPU (Vulkan)
pareto frontier (no model is both faster and more accurate) other models
highlighted rows are recommended picks or on pareto frontier
| 1 | Qwen3-ASR-1.7Baccurate pick | 7.14 | 4.1× | 2.5× | 2.04 GB | Apache-2.0 |
| 2 | Qwen3-ASR-0.6Boverall pick | 7.57 | 9.7× | 5.3× | 0.79 GB | Apache-2.0 |
| 3 | whisper-large-v3 | 7.98 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 4 | Breeze-ASR-25 | 8.10 * | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 5 | whisper-large-v3-turbo | 8.50 | 3.1× | 1.4× | 0.83 GB | Apache-2.0 |
| 6 | Fun-ASR-Nano-2512fast pick | 8.59 | 10.7× | 7.9× | 0.83 GB | FunASR Model Open Source License Agreement v1.1 |
| 7 | Fun-ASR-MLT-Nano-2512 | 8.64 | 8.8× | 5.5× | 0.83 GB | FunASR Model Open Source License Agreement v1.1 |
| 8 | MOSS-Transcribe-Diarize | 9.23 | 3.1× | 1.6× | 0.92 GB | Apache-2.0 |
| 9 | SenseVoiceSmall | 10.12 | 27.7× | 19.7× | 0.24 GB | model-license (FunASR MODEL_LICENSE) |
| 10 | Voxtral-Mini-4B-Realtime-2602 | 10.41 | 0.8× | 0.6× | 4.41 GB | Apache-2.0 |
| 11 | cohere-transcribe-03-2026 | 11.18 | 8.1× | 4.3× | 2.25 GB | Apache-2.0 |
| 12 | whisper-medium | 13.13 | 4.1× | 1.9× | 0.77 GB | Apache-2.0 |
| 13 | moonshine-tiny-zh | 13.88 | 42.9× | 28.4× | 0.03 GB | MIT |
| 14 | whisper-large-v2 | 15.39 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 15 | moonshine-base-zh | 17.79 | 28.0× | 18.3× | 0.07 GB | MIT |
| 16 | nemotron-3.5-asr-streaming-0.6b | 18.87 | 17.1× | 10.6× | 0.70 GB | OpenMDW-1.1 |
| 17 | whisper-large | 19.32 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 18 | whisper-small | 23.06 | 11.7× | 5.9× | 0.25 GB | Apache-2.0 |
| 19 | whisper-base | 36.21 | 29.0× | 19.3× | 0.08 GB | Apache-2.0 |
| 20 | whisper-tiny | 41.91 | 43.1× | 39.8× | 0.04 GB | Apache-2.0 |