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 · 12 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 whisper-large-v3most accurate 4.694.15–5.26 2.1× 0.6× 1.55 GB Apache-2.0
2 whisper-large-v3-turbo 5.815.26–6.45 3.4× 0.8× 0.83 GB Apache-2.0
3 cohere-transcribe-03-2026! 6.155.51–6.85 8.0× 3.0× 2.41 GB Apache-2.0
4 canary-1b-v2!overall pickfast pick 6.886.32–7.44 13.2× 6.7× 1.10 GB CC-BY-4.0
5 parakeet-tdt-0.6b-v3 7.376.75–7.97 12.5× 7.5× 0.72 GB CC-BY-4.0
6 parakeet-primeline 8.197.45–9.07 12.5× 7.5× 0.72 GB CC-BY-4.0
7 whisper-medium 8.597.97–9.27 4.3× 1.1× 0.77 GB Apache-2.0
8 Qwen3-ASR-1.7B 12.5011.66–13.28 3.8× 2.0× 2.04 GB Apache-2.0
9 whisper-small 16.8215.87–17.80 12.1× 3.4× 0.25 GB Apache-2.0
10 nemotron-3.5-asr-streaming-0.6b 17.5416.58–18.50 14.5× 7.5× 0.70 GB OpenMDW-1.1
11 Qwen3-ASR-0.6B 25.0624.04–26.12 8.0× 4.3× 0.79 GB Apache-2.0
12 Fun-ASR-MLT-Nano-2512! 59.3457.63–61.19 9.0× 4.5× 0.83 GB FunASR Model Open Source License Agreement v1.1