OpenAI releases Whisper Large v3 Turbo — 8× faster, same accuracy
The new Whisper model matches large-v3 accuracy but runs 8× faster. Transcribes 1 hour of audio in under 4 minutes on a modern laptop CPU.
The new Whisper model matches large-v3 accuracy but runs 8× faster. Transcribes 1 hour of audio in under 4 minutes on a modern laptop CPU.
OpenAI pushed Whisper Large v3 Turbo to the repo last week. The model is a distilled version of large-v3: same vocabulary, same architecture, but compressed so it runs about 8 times faster with minimal accuracy loss. Word error rate on English is within 0.5% of the full large model.
On an M3 MacBook Pro, turbo transcribes 60 minutes of audio in about 3.5 minutes. The full large-v3 takes 28 minutes on the same machine. For the vast majority of use cases — meeting notes, interviews, podcasts — turbo is now the right default.
If you’ve been stuck on `base` or `small` because large was too slow, this is the upgrade. Run `pip install -U openai-whisper` to get it, then swap `--model base` for `--model turbo` in your command.
One command: `pip install -U openai-whisper`. Then change `--model base` to `--model turbo` in whatever script or alias you have. That’s it.
See the full Whisper guide →At 8× base speed, turbo is now fast enough to process a 1-hour meeting recording in about the time it takes to make a coffee. Combine with the Whisper guide’s batch script to auto-transcribe a folder of recordings.
Get tomorrow's best AI project in your email. With a guide that works. Free. No spam.