Whisper Large v3 Turbo
Whisper large-v3 accuracy at roughly 8x the transcription speed.
About
OpenAI's large-v3-turbo (October 2024) keeps the large-v3 encoder but distills the decoder from 32 layers down to 4, shrinking the model to 809M parameters. The result transcribes about 8x faster than large-v3 with minimal accuracy loss, and it is now the default model in the Whisper CLI. MIT licensed and multilingual — unlike the English-only Distil-Whisper family. Honest caveats: it still processes audio in 30-second chunks, so very long files need a chunking pipeline; and for pure-English speed runs, smaller distilled checkpoints can be faster still.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
8x faster than large-v3 with minimal WER degradation (OpenAI).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: October 2024 release; 809M params, 4-layer distilled decoder, ~8x speedup verified via technical writeups.; MIT license (openai/whisper repo).; Multilingual vs English-only distill distinction documented.; 30s chunking limitation disclosed..