whatmodel.sh / stt / Distil-Whisper distil-large-v3

Distil-Whisper distil-large-v3

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Distil-Whisper distil-large-v3 is an open-weights speech-to-text model from Hugging Face (Distil-Whisper), released March 2024. Its published size is 0.76B parameters. Licence: mit.

Catalogue identity notes

Exact checkpoint distil-whisper/distil-large-v3; distinct from catalogue distil-large-v3-5. Params: HF safetensors total. released: HF repo creation month.

55.5±1.7#66 of 162 speech-to-text modelsfrom 15 benchmark results across 3 sources

Score within the Open ASR English short-form WER (March 2026) comparison set only.

$ how it compares

The speech-to-text models ranked either side of Distil-Whisper distil-large-v3, out of 176 with a score, then more from Hugging Face (Distil-Whisper).

#modelscorewhereRAMreleased
75Omnilingual ASR LLM 1B v2 Meta56±3local~6GB2025-12
76Omnilingual ASR LLM 1B Meta56±2local~6GB2025-11
77Distil-Whisper distil-large-v3 Hugging Face (Distil-Whisper)56±2local~2.3GB2024-03
78Omnilingual ASR LLM 3B v2 Meta56±3local~11GB2025-12
79LiteWhisper large-v3 Efficient-Speech (LiteASR)55±3local~3.7GB2025-02
more from Hugging Face (Distil-Whisper)
60Distil-Whisper distil-large-v3.5 Hugging Face (Distil-Whisper)58±1local~2.3GB2025-03
122Distil-Whisper distil-large-v2 Hugging Face (Distil-Whisper)48±2local~2.3GB2023-10
123Distil-Whisper distil-small.en Hugging Face (Distil-Whisper)48±4local~0.9GB2023-12

see all speech-to-text models · find the ones that fit your machine

$ what it's good at

long-form transcription▁▁▄▁·▁▂█▄▁▁▁ #17 of 27 * · 10% of its score · 1 test

Beats Parakeet TDT 1.1B, Parakeet RNNT 1.1B here, though they rank higher overall. Trails Whisper large (v1), Whisper large-v2, Rev AI Machine (Reverb ASR) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Open ASR English long-form WER9.8% ✎▁▁▅▁·▃▅█▆▁▂▂ beats 11/30avg 12%▇▇▇▇ 10%typical
other tests14 not in a task · 89% of its score
testresultvs the others in its rundifficultycountswhy
Open ASR English short-form WER (March 2026)7.5% ✎▁▁▁▃··▁▁▂██▆ beats 46/86avg 8.2%▇▇▇▇ 10%blunt
SPGISpeech test WER lab3.3% ✎▄▄··▄██··▄██ beats 2/11avg 5.3%▇▇▇▇ 10%sharp
Meanwhile long-form WER lab7.0% ✎█·······██·█ beats 1/3avg 11%▇▇▇ 8.7%sharp
TED-LIUM long-form WER lab3.9% ✎█······█·█·█ beats 2/3avg 6.9%▇▇▇ 8.2%sharp
AMI WER (Open ASR / ESB test split) lab15.2% ✎▃█▃▃·····▅·▃ beats 4/8avg 24%▇▇▇ 7.7%typical
Earnings-22 WER (short-form, Open ASR / ESB test split) lab11.8% ✎▃······██▃·▃ beats 2/8avg 18%▇▇▇ 7.3%typical
Rev16 long-form WER lab13.8% ✎█········███ beats 2/3avg 20%▇▇▇ 6.7%typical
GigaSpeech test WER lab10.1% ✎▁····▁··▁▄█▂ beats 9/17avg 15%▇▇▇ 6.4%typical
6 more tests
testresultvs the others in its rundifficultycountswhy
Earnings-21 long-form WER lab10.5% ✎█········███ beats 2/3avg 15%▇▇▇ 6.3%typical
Earnings-22 long-form WER lab15.1% ✎█······█·█·█ beats 1/3avg 20%▇▇ 5.9%typical
LibriSpeech test-other WER lab5.2% ✎▁▁▁▂▅▃▅▄█▂▂▁ beats 20/54avg 6.6%▇▇ 5.4%typical
Distil-Whisper · 4-set out-of-distribution short-form mean WER lab9.7% ✎█·█··█·█···█ beats 3/4avg 10%▇▇ 4.6%typical
LibriSpeech test-clean WER lab2.5% ✎▁·▁·▁▁▁▁▃▇█▆ beats 30/100avg 2.6%▇ 1.6%blunt
TED-LIUM test WER (short-form) lab3.9% ✎▁·▁▇█▃·▁▁▃█▄ beats 21/27avg 5.3%▇ 0.4%blunt

$ where to run it

Other ways to run it

  1. Run it as published · needs ~2.32 GB · from huggingface.co/distil-whisper (official)
    uv run --with transformers --with torch python -c "from transformers import pipeline; print(pipeline('automatic-speech-recognition', model='distil-whisper/distil-large-v3')('audio.wav', return_timestamps=True)['text'])"

Official weights and documentation: huggingface.co/distil-whisper ↗

$ how the score was worked out · distil-whisper/distil-large-v… 74%, Open ASR Leaderboard 20%, huggingface/distil-whisper RE… 5%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
distil-whisper/distil-large-v3.5 model card · lab-reported12▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 74%MIT
Open ASR Leaderboard2▇▇▇▇▇▇ 20%Apache-2.0
huggingface/distil-whisper README (Distil-Whisper headline WER table) · lab-reported1▇ 5%MIT
prior · pull toward the middle▇ 1%
weighted average= 55.5

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json