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

Greek · 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 cohere-transcribe-03-2026!overall pickmost accuratefast pick 8.968.23–9.73 8.0× 3.0× 2.41 GB Apache-2.0
2 whisper-large-v3 11.5310.70–12.34 2.1× 0.6× 1.55 GB Apache-2.0
3 whisper-large-v3-turbo 13.2612.35–14.13 3.4× 0.8× 0.83 GB Apache-2.0
4 whisper-medium 20.0619.06–21.13 4.3× 1.1× 0.77 GB Apache-2.0
5 canary-1b-v2! 26.0225.01–26.98 13.2× 6.7× 1.10 GB CC-BY-4.0
6 Qwen3-ASR-1.7B 29.2227.87–30.64 3.8× 2.0× 2.04 GB Apache-2.0
7 whisper-small 33.9832.74–35.34 12.1× 3.4× 0.25 GB Apache-2.0
8 parakeet-primeline 34.7633.60–35.92 12.5× 7.5× 0.72 GB CC-BY-4.0
9 parakeet-tdt-0.6b-v3 35.3334.07–36.55 12.5× 7.5× 0.72 GB CC-BY-4.0
10 Qwen3-ASR-0.6B 49.1247.74–50.55 8.0× 4.3× 0.79 GB Apache-2.0
11 Fun-ASR-MLT-Nano-2512! 103.55101.64–105.60 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1