handy models

speech-to-text benchmarks for transcribe.cpp, the engine behind Handy

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

Czech · 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 canary-1b-v2!overall pickmost accuratefast pick 8.567.91–9.23 13.2× 6.7× 1.10 GB CC-BY-4.0
2 whisper-large-v3 10.509.79–11.22 2.1× 0.6× 1.55 GB Apache-2.0
3 whisper-large-v3-turbo 11.8111.06–12.54 3.4× 0.8× 0.83 GB Apache-2.0
4 parakeet-tdt-0.6b-v3 12.3111.42–13.18 12.5× 7.5× 0.72 GB CC-BY-4.0
5 parakeet-primeline 15.0514.18–16.05 12.5× 7.5× 0.72 GB CC-BY-4.0
6 whisper-medium 22.8521.93–23.89 4.3× 1.1× 0.77 GB Apache-2.0
7 Qwen3-ASR-1.7B 22.9722.00–24.09 3.8× 2.0× 2.04 GB Apache-2.0
8 nemotron-3.5-asr-streaming-0.6b 23.0021.89–24.02 14.5× 7.5× 0.70 GB OpenMDW-1.1
9 whisper-small 40.5739.32–41.83 12.1× 3.4× 0.25 GB Apache-2.0
10 Qwen3-ASR-0.6B 44.5043.23–45.72 8.0× 4.3× 0.79 GB Apache-2.0
11 Fun-ASR-MLT-Nano-2512! 53.5651.65–55.84 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1