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 CER — lower is better. Speed is × realtime on Handy target hardware.

accuracy vs speed

Korean · 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 Qwen3-ASR-1.7Baccurate pick 4.60 4.1× 2.5× 2.04 GB Apache-2.0
2 whisper-large-v3 4.89 2.0× 1.0× 1.55 GB Apache-2.0
3 whisper-large-v2 4.99 2.0× 1.0× 1.55 GB Apache-2.0
4 Fun-ASR-MLT-Nano-2512!overall pick 5.20 8.8× 5.5× 0.83 GB FunASR Model Open Source License Agreement v1.1
5 whisper-large-v3-turbo 5.24 3.1× 1.4× 0.83 GB Apache-2.0
6 whisper-large 5.25 2.0× 1.0× 1.55 GB Apache-2.0
7 Voxtral-Mini-4B-Realtime-2602 5.27 0.8× 0.6× 4.41 GB Apache-2.0
8 whisper-medium 5.46 4.1× 1.9× 0.77 GB Apache-2.0
9 Qwen3-ASR-0.6B 5.82 9.7× 5.3× 0.79 GB Apache-2.0
10 cohere-transcribe-03-2026! 6.57 8.1× 4.3× 2.25 GB Apache-2.0
11 whisper-small 7.70 11.7× 5.9× 0.25 GB Apache-2.0
12 moonshine-base-kofast pick 8.12 17.5× 11.2× 0.07 GB MIT
13 SenseVoiceSmall 8.27 27.7× 19.7× 0.24 GB model-license (FunASR MODEL_LICENSE)
14 nemotron-3.5-asr-streaming-0.6b 8.89 17.1× 10.6× 0.70 GB OpenMDW-1.1
15 moonshine-tiny-ko 9.00 30.3× 20.5× 0.03 GB MIT
16 whisper-base 12.98 29.0× 19.3× 0.08 GB Apache-2.0
17 whisper-tiny 19.07 43.1× 39.8× 0.04 GB Apache-2.0