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

Maltese · 8 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 canary-1b-v2!overall pickmost accuratefast pick 19.7517.99–22.38 13.2× 6.7× 1.10 GB CC-BY-4.0
2 parakeet-tdt-0.6b-v3 20.6319.86–21.43 12.5× 7.5× 0.72 GB CC-BY-4.0
3 parakeet-primeline 24.7423.87–25.63 12.5× 7.5× 0.72 GB CC-BY-4.0
4 whisper-large-v3 68.7967.66–70.06 2.1× 0.6× 1.55 GB Apache-2.0
5 whisper-large-v3-turbo 70.9269.87–71.96 3.4× 0.8× 0.83 GB Apache-2.0
6 whisper-medium 85.0684.12–86.01 4.3× 1.1× 0.77 GB Apache-2.0
7 Fun-ASR-MLT-Nano-2512! 91.7190.91–92.56 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1
8 whisper-small 97.3295.64–99.20 12.1× 3.4× 0.25 GB Apache-2.0