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 · 17 models

AMD Ryzen 4750U GPU (Vulkan)
pareto frontier (no model is both faster and more accurate) other models
highlighted rows are recommended picks or on pareto frontier
1 cohere-transcribe-03-2026!overall pickaccurate pickfast pick 8.96 8.1× 4.3× 2.25 GB Apache-2.0
2 whisper-large-v3 11.53 2.0× 1.0× 1.55 GB Apache-2.0
3 whisper-large-v3-turbo 13.26 3.1× 1.4× 0.83 GB Apache-2.0
4 whisper-large-v2 13.51 2.0× 1.0× 1.55 GB Apache-2.0
5 whisper-large 18.79 2.0× 1.0× 1.55 GB Apache-2.0
6 whisper-medium 20.06 4.1× 1.9× 0.77 GB Apache-2.0
7 canary-1b-v2! 26.02 14.1× 8.3× 1.07 GB CC-BY-4.0
8 Qwen3-ASR-1.7B 29.22 4.1× 2.5× 2.04 GB Apache-2.0
9 parakeet-ultra 32.25 23.0× 13.3× 0.69 GB CC-BY-4.0
10 orukeet 33.12 25.1× 13.6× 0.69 GB CC-BY-SA-4.0
11 whisper-small 33.98 11.7× 5.9× 0.25 GB Apache-2.0
12 parakeet-primeline 34.76 25.0× 13.6× 0.69 GB CC-BY-4.0
13 parakeet-tdt-0.6b-v3 35.33 25.1× 13.7× 0.69 GB CC-BY-4.0
14 Qwen3-ASR-0.6B 49.12 9.7× 5.3× 0.79 GB Apache-2.0
15 whisper-base 59.14 29.0× 19.3× 0.08 GB Apache-2.0
16 whisper-tiny 77.76 43.1× 39.8× 0.04 GB Apache-2.0
17 Fun-ASR-MLT-Nano-2512! 103.55 8.8× 5.5× 0.83 GB FunASR Model Open Source License Agreement v1.1