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.
No supported recommendation: the lowest error is at least 30% and its 95% confidence interval does not overlap 30%.
accuracy vs speed
Occitan · 8 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 | whisper-large-v3 | 69.15 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 2 | whisper-large-v3-turbo | 70.94 | 3.1× | 1.4× | 0.83 GB | Apache-2.0 |
| 3 | whisper-large-v2 | 75.41 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 4 | whisper-large | 76.72 | 2.0× | 1.0× | 1.55 GB | Apache-2.0 |
| 5 | whisper-medium | 81.55 | 4.1× | 1.9× | 0.77 GB | Apache-2.0 |
| 6 | whisper-base | 88.62 | 29.0× | 19.3× | 0.08 GB | Apache-2.0 |
| 7 | whisper-small | 90.43 | 11.7× | 5.9× | 0.25 GB | Apache-2.0 |
| 8 | whisper-tiny | 96.44 | 43.1× | 39.8× | 0.04 GB | Apache-2.0 |