Public benchmark · launching soon

The live clinical-transcription benchmark launches here.

A neutral, head-to-head benchmark of speech-to-text models on a public clinical corpus, scored on normalized word error rate, medical-term F1, and cost per minute. Designed to re-run on demand, with receipts attached; latest run shown below.

Every published number will carry a receipt: the audio, the reference transcript, and each model's output, checkable by anyone. No vendor's thumb on the scale.