All benchmarksTest WER (lower is better)
LibriSpeech (clean), Test WER
As the model cards publish them, each model’s best result. Facthory has not run them; results are self-reported unless the hub verified them.
| # | Model | Result | Source |
|---|---|---|---|
| 1 | whisper-medium openai · test | 2.9 | Model card |
| 2 | whisper-large openai · test | 3 | Model card |
| 3 | whisper-small openai · test | 3.43 | Model card |
| 4 | whisper-medium.en openai · test | 4.12 | Model card |
| 5 | whisper-base openai · test | 5.01 | Model card |
| 6 | whisper-tiny openai · test | 7.54 | Model card |
| 7 | whisper-tiny.en openai · test | 8.44 | Model card |
