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

French · 27 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 Voxtral-Small-24B-2507 3.86 — — 24.04 GB Apache-2.0
2 Voxtral-Mini-3B-2507 4.51 1.1× 0.6× 4.66 GB Apache-2.0
3 Qwen3-ASR-1.7Baccurate pick 4.52 4.1× 2.5× 2.04 GB Apache-2.0
4 parakeet-ultraoverall pickfast pick 4.66 23.0× 13.3× 0.69 GB CC-BY-4.0
5 canary-1b-v2! 5.09 14.1× 8.3× 1.07 GB CC-BY-4.0
6 cohere-transcribe-03-2026! 5.23 8.1× 4.3× 2.25 GB Apache-2.0
7 parakeet-tdt-0.6b-v3 5.30 25.1× 13.7× 0.69 GB CC-BY-4.0
8 whisper-large-v3 5.39 2.0× 1.0× 1.55 GB Apache-2.0
9 whisper-large-v3-turbo 5.51 3.1× 1.4× 0.83 GB Apache-2.0
10 whisper-large-v2 5.81 2.0× 1.0× 1.55 GB Apache-2.0
11 orukeet 6.17 25.1× 13.6× 0.69 GB CC-BY-SA-4.0
12 Voxtral-Mini-4B-Realtime-2602 6.29 0.8× 0.6× 4.41 GB Apache-2.0
13 parakeet-primeline 6.35 25.0× 13.6× 0.69 GB CC-BY-4.0
14 granite-speech-4.1-2b-nar! 6.76 4.0× 2.1× 2.33 GB Apache-2.0
15 whisper-large 6.95 2.0× 1.0× 1.55 GB Apache-2.0
16 canary-1b-flash! 7.22 15.5× 8.8× 0.98 GB CC-BY-4.0
17 canary-1b! 7.44 9.8× 6.5× 1.08 GB CC-BY-NC-4.0
18 granite-speech-4.1-2b! 7.61 4.6× 2.8× 2.38 GB Apache-2.0
19 Qwen3-ASR-0.6B 7.76 9.7× 5.3× 0.79 GB Apache-2.0
20 whisper-medium 8.07 4.1× 1.9× 0.77 GB Apache-2.0
21 granite-4.0-1b-speech! 8.44 4.6× 2.8× 2.38 GB Apache-2.0
22 canary-180m-flash! 8.53 33.1× 21.7× 0.20 GB CC-BY-4.0
23 granite-speech-4.1-2b-plus! 8.82 4.6× 2.6× 2.18 GB Apache-2.0
24 nemotron-3.5-asr-streaming-0.6b 10.78 17.1× 10.6× 0.70 GB OpenMDW-1.1
25 whisper-small 13.30 11.7× 5.9× 0.25 GB Apache-2.0
26 whisper-base 27.91 29.0× 19.3× 0.08 GB Apache-2.0
27 whisper-tiny 44.19 43.1× 39.8× 0.04 GB Apache-2.0