best model by language

Benchmarked on the FLEURS test split at Q8_0. Error is WER — lower is better. Speed is × realtime on Handy target hardware.

accuracy vs speed

Swedish · 11 models · left is slower, down is better · log speed axis

AMD Ryzen 4750U GPU (Vulkan)
pareto frontier (no model is both faster and more accurate) other models
1 whisper-large-v3most accurate 7.807.22–8.37 2.1× 0.6× 1.55 GB Apache-2.0
2 whisper-large-v3-turbo 8.728.12–9.34 3.4× 0.8× 0.83 GB Apache-2.0
3 canary-1b-v2!overall pickfast pick 9.749.06–10.44 13.2× 6.7× 1.10 GB CC-BY-4.0
4 whisper-medium 12.4711.70–13.21 4.3× 1.1× 0.77 GB Apache-2.0
5 parakeet-tdt-0.6b-v3 15.2514.38–16.05 12.5× 7.5× 0.72 GB CC-BY-4.0
6 parakeet-primeline 16.4215.53–17.29 12.5× 7.5× 0.72 GB CC-BY-4.0
7 Qwen3-ASR-1.7B 19.6818.70–20.62 3.8× 2.0× 2.04 GB Apache-2.0
8 whisper-small 23.1021.94–24.32 12.1× 3.4× 0.25 GB Apache-2.0
9 nemotron-3.5-asr-streaming-0.6b 24.3223.32–25.30 14.5× 7.5× 0.70 GB OpenMDW-1.1
10 Qwen3-ASR-0.6B 35.7234.53–36.94 8.0× 4.3× 0.79 GB Apache-2.0
11 Fun-ASR-MLT-Nano-2512! 75.3672.22–79.42 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1