Results
OpenAI GPT Transcribe accuracy and speed
OpenAI's hosted transcription API. We measure it the same way as everything else here, over the same audio, so the comparison against our own cloud model is like for like.
80
Blended score
7.8%
Word error rate
0.78 s
Wait for the text
Macro average over 8 datasets
| Dataset | Score ↑ | WER ↓ | Speed ↑ | Recall ↑ | F-score ↑ | Response ↓ |
|---|---|---|---|---|---|---|
| AMI Meetings20260730-161636, apples-general-esb-20260811 | 75 | 10.0% | 15× | — | — | 0.78 s |
| Common Voice, Australian Englishapples-general-public-20260811 | 92 | 3.4% | 8.3× | — | — | 0.78 s |
| Common Voice, spontaneous Englishapples-general-public-20260811 | 78 | 8.6% | 14× | — | — | 0.78 s |
| Earnings 2220260730-161636, apples-general-esb-20260811 | 79 | 8.4% | 19× | — | — | 0.78 s |
| Earnings 22 (with vocabulary)20260731-175836 | 63 | 14.8% | 18× | 90.4% | 93.2% | 0.78 s |
| LibriSpeech Other20260730-161636, apples-general-esb-20260811 | 92 | 3.3% | 16× | — | — | 0.78 s |
| Loquaciousapples-general-public-20260811 | 85 | 6.0% | 12× | — | — | 0.78 s |
| Phonetic vocabulary20260731-175836 | 80 | 8.1% | 4.7× | 91.4% | 94.1% | 0.78 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