Whisper
AI Model
Speech & Music
Current
The open speech-recognition model the whole industry builds on.
You're looking at an AI model record: who built it, what it can do, and the hardware and data behind it.
Whisper (September 2022, MIT license) brought robust multilingual transcription to anyone with a GPU — 680,000 hours of training audio distilled into checkpoints from 39M to 1.5B parameters. It remains the default open ASR in products everywhere, from subtitling tools to call centers.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Developed by
Sources
Record reviewed as of 2026-01-15 — see Methodology.