Distil Large v3.5
Distil-Whisper · released Dec 5, 2024 · distil-whisper/distil-large-v3.5
Speech to textTranscribes a recording into words
- Type
- Open weightsMIT License
- Languages
- 1
- Size
- 0.8B
Context measured in tokens
Our take
Written Sep 5, 2026Distil Large v3.5 is a tiny English-only speech-to-text model built for speed. At 0.8 billion parameters and under an MIT licence, it fits on almost any device and processes an hour of audio in about four seconds.
Pick this when you need fast, free local transcription of podcasts, video or accented speech — it handles those better than most models. Use it for commercial projects that need a permissive licence, or anywhere you cannot send audio to a cloud service. Skip it if you need clean read-aloud accuracy, meeting-room transcription, or any language other than English.
The case for it
- Processes audio at 874 times real time — an hour in roughly four seconds on the benchmark rig.
- Better than most on podcasts and video, with an 8.2% word error rate against a worst-in-field of 17.2%.
- Handles accented speech better than most: 10.8% on earnings-call accents and 2.5% on European-accented English.
- Tiny 0.8B parameter count and MIT licence make it free for commercial use and easy to deploy locally.
The case against it
- Mediocre overall accuracy at 6.1% word errors — roughly one word in sixteen wrong, behind the field best of 4.4%.
- Weak on clean read-aloud audio at 1.9%, nearly double the field median of 1.4%.
- Struggles with meeting recordings at 12.1%, worse than the field median of 10.6%.
How good is it?
An open transcription model for turning speech into text, though it struggles with clear read-aloud recordings.
- transcribing clear recordings of people reading aloudClean read speech · 73rd of 92
TranscriptionTurning speech into text3 of 5Open ASR WER · 46th of 76
94.6%
Misses roughly one word in 19, averaged over nine English test sets.
879×33rd of 74
an hour of audio in 4 seconds, on the board's own hardware. Your machine will differ.
1
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 21.1 GB spare means a 10% error in the size would not change the answer.
Apple M1 Pro (16-core GPU) · 32 GB
Room to spare. 22.3 GB spare means a 10% error in the size would not change the answer.
Apple M2 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 4.3 GB spare means a 10% error in the size would not change the answer.
These cards answer whether Distil Large v3.5 loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsMIT License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
MIT License
Fully permissive: do anything with attribution. No patent grant, unlike Apache-2.0.
Identifiers
- Hugging Face
- distil-whisper/distil-large-v3.5
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- distil-whisper-distil-large-v3-5