hjLabs AI Playground
Starting the on-device runtime…
whisper-tiny.en is about 40 MB and transcribes faster than real time on almost anything, but it stumbles on accents, crosstalk and technical vocabulary. whisper-base roughly doubles the size and is the best general default. whisper-small is noticeably more accurate and noticeably slower, and is about the practical ceiling in a browser tab. distil-whisper trades a little accuracy for a large speed gain. This page compares them on download size, warm-up time, throughput and multilingual coverage so you can pick once and stop guessing.
whisper tiny vs base vs smallwhisper model comparisonbest whisper model browserdistil-whisper vs whisperwhisper size accuracywhisper onnx webwhisper.en vs multilingual
English-only variants. They are the same size but more accurate on English than the multilingual model of the same size, because none of the capacity is spent on other languages.
For English at scale, usually yes — it is several times faster than the equivalent Whisper for a small accuracy cost. For multilingual audio, stay with standard Whisper.