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

English · 17 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 Qwen3-ASR-1.7Bmost accurate 3.232.85–3.65 3.8× 2.0× 2.04 GB Apache-2.0
2 parakeet-unified-en-0.6boverall pickfast pick 3.993.60–4.42 12.5× 7.5× 0.71 GB CC-BY-4.0
3 whisper-large-v3 4.033.59–4.46 2.1× 0.6× 1.55 GB Apache-2.0
4 Breeze-ASR-25 4.143.70–4.61 2.1× 0.6× 1.67 GB Apache-2.0
5 Qwen3-ASR-0.6B 4.233.76–4.69 8.0× 4.3× 0.79 GB Apache-2.0
6 whisper-large-v3-turbo 4.383.95–4.84 3.4× 0.8× 0.83 GB Apache-2.0
7 canary-1b-v2! 4.474.03–4.95 13.2× 6.7× 1.10 GB CC-BY-4.0
8 whisper-medium 4.644.20–5.15 4.3× 1.1× 0.77 GB Apache-2.0
9 parakeet-primeline 4.824.39–5.34 12.5× 7.5× 0.72 GB CC-BY-4.0
10 parakeet-tdt-0.6b-v3 4.834.39–5.30 12.5× 7.5× 0.72 GB CC-BY-4.0
11 Fun-ASR-MLT-Nano-2512! 4.904.40–5.48 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1
12 cohere-transcribe-03-2026! 5.084.56–5.59 8.0× 3.0× 2.41 GB Apache-2.0
13 canary-180m-flash! 5.985.26–6.78 31.9× 21.3× 0.20 GB CC-BY-4.0
14 whisper-small 6.515.88–7.22 12.1× 3.4× 0.25 GB Apache-2.0
15 SenseVoiceSmall 7.146.54–7.77 32.5× 15.5× 0.24 GB model-license (FunASR MODEL_LICENSE)
16 nemotron-3.5-asr-streaming-0.6b 7.907.23–8.54 14.5× 7.5× 0.70 GB OpenMDW-1.1
17 Voxtral-Mini-4B-Realtime-2602 11.7710.02–13.71 0.9× 0.6× 4.73 GB Apache-2.0