Whisper Large v3 Turbo
OpenAI · released Oct 1, 2024 · openai/whisper-large-v3-turbo
Speech to textTranscribes a recording into words
- Type
- Open weightsMIT License
- Languages
- 99
- Size
- 0.8B
Context measured in tokens
Our take
Written Sep 11, 2026Whisper Large v3 Turbo is a tiny, MIT-licensed speech-to-text model built for speed over accuracy. It transcribes audio in 99 languages at roughly 792 times real time, but sits below average on every English recording condition we track.
Pick this when raw speed is the overriding concern, or when you need broad language coverage with a permissive licence on minimal hardware. Use it for clean, scripted audio where even a below-average model is good enough. Skip it if accuracy matters on meetings, accented speech, or podcasts, or if you need measured quality in any language other than English.
The case for it
- Extremely fast: roughly 792 times real time on benchmark hardware, turning an hour of audio into text in about five seconds.
- MIT licence with a 0.8-billion-parameter footprint, allowing fine-tuning, redistribution and commercial use.
- 99 languages supported, including English, Chinese, German, Spanish, Russian, Korean, French and Japanese.
The case against it
- Below-average accuracy on every measured English condition, with a 6.36% overall word error rate against a 5.1% median across 62 models.
- Particularly weak on meeting-room audio: 13.88% word error rate, only 2.3 percentage points above the worst measured.
- Struggles with accented speech and podcasts, trailing the median on both.
How good is it?
An open speech-to-text model for turning recordings into written text, though it trails most models on accuracy.
- turning spoken English into written textOpen ASR WER · 61st of 76
- transcribing recordings of meetings in a roomRecorded meetings · 76th of 92
- transcribing clear recordings of people reading aloudClean read speech · 79th of 92
TranscriptionTurning speech into text2.5 of 5Open ASR WER · 61st of 76
93.6%
Misses roughly one word in 16, averaged over nine English test sets.
797×37th of 74
an hour of audio in 5 seconds, on the board's own hardware. Your machine will differ.
99
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.
Which languages ↓ ↑
English · Chinese · German · Spanish · Russian · Korean · French · Japanese · Portuguese · Turkish · Polish · Catalan · Dutch · Arabic · Swedish · Italian · Indonesian · Hindi · Finnish · Vietnamese · Hebrew · Ukrainian · Greek · Malay · Czech · Romanian · Danish · Hungarian · Tamil · Norwegian · Thai · Urdu · Croatian · Bulgarian · Lithuanian · Latin · Māori · Malayalam · Welsh · Slovak · Telugu · Persian · Latvian · Bangla · Serbian · Azerbaijani · Slovenian · Kannada · Estonian · Macedonian · Breton · Basque · Icelandic · Armenian · Nepali · Mongolian · Bosnian · Kazakh · Albanian · Swahili · Galician · Marathi · Punjabi · Sinhala · Khmer · Shona · Yoruba · Somali · Afrikaans · Occitan · Georgian · Belarusian · Tajik · Sindhi · Gujarati · Amharic · Yiddish · Lao · Uzbek · Faroese · Haitian Creole · Pashto · Turkmen · Norwegian Nynorsk · Maltese · Sanskrit · Luxembourgish · Burmese · bo · Filipino · Malagasy · Assamese · Tatar · Hawaiian · Lingala · Hausa · ba · Javanese · Sundanese
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 Whisper Large v3 Turbo 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.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked 4 hours ago — each listing carries its own date.
- per minute of audio
- $0.001
- Context served
- —
- Throughput
- Not measured
| Provider | Price per minute of audio | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| GroqDirect | $0.001checked 4 hours ago | not reported | not measured | Unknown | Unknown | Unknown |
| IBM watsonxDirect | $0.006checked 4 hours ago | not reported | not measured | Unknown | Unknown | Unknown |
Across the 2 listings we hold: 0 say they do not train on prompts, 0 say they do and 2 do not say. 0 appear in the zero-retention registry we check; the rest are unknown to us.
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.
- 2 of 2 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 2 listings do not say whether they train on prompts.
- We hold no cached-input rate for any of its listings.
- We hold no batch or off-peak rate for any of its listings.
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
- openai/whisper-large-v3-turbo
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- openai-whisper-large-v3-turbo