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

Hungarian · 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 canary-1b-v2!overall pickaccurate pickfast pick 13.06 14.1× 8.3× 1.07 GB CC-BY-4.0
2 parakeet-ultra 13.09 23.0× 13.3× 0.69 GB CC-BY-4.0
3 whisper-large-v3 13.40 2.0× 1.0× 1.55 GB Apache-2.0
4 orukeet 13.53 25.1× 13.6× 0.69 GB CC-BY-SA-4.0
5 whisper-large-v3-turbo 15.07 3.1× 1.4× 0.83 GB Apache-2.0
6 parakeet-tdt-0.6b-v3 16.06 25.1× 13.7× 0.69 GB CC-BY-4.0
7 parakeet-primeline 17.52 25.0× 13.6× 0.69 GB CC-BY-4.0
8 whisper-large-v2 17.84 2.0× 1.0× 1.55 GB Apache-2.0
9 whisper-large 21.97 2.0× 1.0× 1.55 GB Apache-2.0
10 whisper-medium 26.07 4.1× 1.9× 0.77 GB Apache-2.0
11 nemotron-3.5-asr-streaming-0.6b 32.12 17.1× 10.6× 0.70 GB OpenMDW-1.1
12 Qwen3-ASR-1.7B 32.84 4.1× 2.5× 2.04 GB Apache-2.0
13 whisper-small 42.39 11.7× 5.9× 0.25 GB Apache-2.0
14 Qwen3-ASR-0.6B 56.24 9.7× 5.3× 0.79 GB Apache-2.0
15 whisper-base 72.26 29.0× 19.3× 0.08 GB Apache-2.0
16 whisper-tiny 90.54 43.1× 39.8× 0.04 GB Apache-2.0
17 Fun-ASR-MLT-Nano-2512! 113.21 8.8× 5.5× 0.83 GB FunASR Model Open Source License Agreement v1.1