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

Danish · 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 11.3410.73–11.93 13.2× 6.7× 1.10 GB CC-BY-4.0
2 whisper-large-v3 12.4811.78–13.16 2.1× 0.6× 1.55 GB Apache-2.0
3 whisper-large-v3-turbo 13.6012.93–14.30 3.4× 0.8× 0.83 GB Apache-2.0
4 parakeet-tdt-0.6b-v3 18.6417.84–19.43 12.5× 7.5× 0.72 GB CC-BY-4.0
5 parakeet-primeline 20.5219.78–21.34 12.5× 7.5× 0.72 GB CC-BY-4.0
6 whisper-medium 21.1720.35–22.01 4.3× 1.1× 0.77 GB Apache-2.0
7 Qwen3-ASR-1.7B 21.2320.40–22.14 3.8× 2.0× 2.04 GB Apache-2.0
8 nemotron-3.5-asr-streaming-0.6b 28.5127.49–29.47 14.5× 7.5× 0.70 GB OpenMDW-1.1
9 whisper-small 35.5234.52–36.61 12.1× 3.4× 0.25 GB Apache-2.0
10 Qwen3-ASR-0.6B 36.0735.05–37.08 8.0× 4.3× 0.79 GB Apache-2.0
11 Fun-ASR-MLT-Nano-2512! 69.9368.45–71.45 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1