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

Polish · 18 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 whisper-large-v3 4.69 2.0× 1.0× 1.55 GB Apache-2.0
2 whisper-large-v3-turboaccurate pick 5.81 3.1× 1.4× 0.83 GB Apache-2.0
3 whisper-large-v2 5.87 2.0× 1.0× 1.55 GB Apache-2.0
4 cohere-transcribe-03-2026! 6.15 8.1× 4.3× 2.25 GB Apache-2.0
5 parakeet-ultraoverall pickfast pick 6.16 23.0× 13.3× 0.69 GB CC-BY-4.0
6 canary-1b-v2! 6.88 14.1× 8.3× 1.07 GB CC-BY-4.0
7 orukeet 7.35 25.1× 13.6× 0.69 GB CC-BY-SA-4.0
8 parakeet-tdt-0.6b-v3 7.37 25.1× 13.7× 0.69 GB CC-BY-4.0
9 whisper-large 7.44 2.0× 1.0× 1.55 GB Apache-2.0
10 parakeet-primeline 8.19 25.0× 13.6× 0.69 GB CC-BY-4.0
11 whisper-medium 8.59 4.1× 1.9× 0.77 GB Apache-2.0
12 Qwen3-ASR-1.7B 12.50 4.1× 2.5× 2.04 GB Apache-2.0
13 whisper-small 16.82 11.7× 5.9× 0.25 GB Apache-2.0
14 nemotron-3.5-asr-streaming-0.6b 17.54 17.1× 10.6× 0.70 GB OpenMDW-1.1
15 Qwen3-ASR-0.6B 25.06 9.7× 5.3× 0.79 GB Apache-2.0
16 whisper-base 35.68 29.0× 19.3× 0.08 GB Apache-2.0
17 whisper-tiny 54.22 43.1× 39.8× 0.04 GB Apache-2.0
18 Fun-ASR-MLT-Nano-2512! 59.34 8.8× 5.5× 0.83 GB FunASR Model Open Source License Agreement v1.1