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

Italian · 20 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-Mini-4B-Realtime-2602 2.25 0.8× 0.6× 4.41 GB Apache-2.0
2 parakeet-ultraoverall pickaccurate pickfast pick 2.34 23.0× 13.3× 0.69 GB CC-BY-4.0
3 whisper-large-v3 2.54 2.0× 1.0× 1.55 GB Apache-2.0
4 Voxtral-Mini-3B-2507 2.56 1.1× 0.6× 4.66 GB Apache-2.0
5 Qwen3-ASR-1.7B 2.68 4.1× 2.5× 2.04 GB Apache-2.0
6 Voxtral-Small-24B-2507 2.69 — — 24.04 GB Apache-2.0
7 whisper-large-v3-turbo 2.77 3.1× 1.4× 0.83 GB Apache-2.0
8 parakeet-tdt-0.6b-v3 3.02 25.1× 13.7× 0.69 GB CC-BY-4.0
9 canary-1b-v2! 3.10 14.1× 8.3× 1.07 GB CC-BY-4.0
10 orukeet 3.11 25.1× 13.6× 0.69 GB CC-BY-SA-4.0
11 parakeet-primeline 3.17 25.0× 13.6× 0.69 GB CC-BY-4.0
12 cohere-transcribe-03-2026! 3.24 8.1× 4.3× 2.25 GB Apache-2.0
13 whisper-large-v2 3.59 2.0× 1.0× 1.55 GB Apache-2.0
14 whisper-large 3.67 2.0× 1.0× 1.55 GB Apache-2.0
15 whisper-medium 4.17 4.1× 1.9× 0.77 GB Apache-2.0
16 Qwen3-ASR-0.6B 5.19 9.7× 5.3× 0.79 GB Apache-2.0
17 nemotron-3.5-asr-streaming-0.6b 5.78 17.1× 10.6× 0.70 GB OpenMDW-1.1
18 whisper-small 7.97 11.7× 5.9× 0.25 GB Apache-2.0
19 whisper-base 17.26 29.0× 19.3× 0.08 GB Apache-2.0
20 whisper-tiny 31.24 43.1× 39.8× 0.04 GB Apache-2.0