Results
Superwhisper S1 Voice accuracy and speed
S1 Voice is our own speech model, trained in house and hosted by us. It is the research line behind what eventually ships in the app, so its numbers move more than the rest of this table.
83
Blended score
6.8%
Word error rate
0.32 s
Wait for the text
Macro average over 8 datasets
| Dataset | Score ↑ | WER ↓ | Speed ↑ | Recall ↑ | F-score ↑ | Response ↓ |
|---|---|---|---|---|---|---|
| AMI Meetingss1-pnc-production-full-esb-20260811 | 81 | 7.5% | 34× | — | — | 0.32 s |
| Common Voice, Australian Englishs1-pnc-production-full-public-20260811 | 94 | 2.4% | 23× | — | — | 0.32 s |
| Common Voice, spontaneous Englishs1-pnc-production-full-public-20260811 | 74 | 10.6% | 44× | — | — | 0.32 s |
| Earnings 22s1-pnc-production-full-esb-20260811 | 79 | 8.3% | 46× | — | — | 0.32 s |
| Earnings 22 (with vocabulary)s1-pnc-production-full-public-20260811 | 69 | 12.4% | 36× | 86.7% | 91.6% | 0.32 s |
| LibriSpeech Others1-pnc-production-full-esb-20260811 | 95 | 2.2% | 38× | — | — | 0.32 s |
| Loquaciouss1-pnc-production-full-public-20260811 | 87 | 5.2% | 27× | — | — | 0.32 s |
| Phonetic vocabularys1-pnc-production-full-public-20260811 | 85 | 5.9% | 10.0× | 62.9% | 68.8% | 0.32 s |
Cloud results do not depend on the machine that made the request, so they are not split by device.
Updated August 27, 2026 from bench commit 5344cab. The second line of each row is the run that produced it.
Method
How this was measured
Cloud results come from calling the model over its public API with the same audio as everything else in the benchmark.
Word error rate counts insertions, deletions and substitutions against a human transcript, so lower is better. Speed is a multiple of real time, so higher is better. The full method and every other model is on the benchmarks page.
Compare
Other models in the benchmark
Same audio, same machines, same method.
Support