Whisper medium
Whisper medium is a 769M-parameter speech-to-text model from OpenAI, released under the MIT license. The reviewed source lists 99+ languages (100 explicitly enumerated in our record), and the model is designed for batch transcription, with live use only through a runtime or chunking wrapper. Among the common setups we test, it ranks highest for subtitles with timing (#1 of 26 in the matcher).
Accuracy-oriented multilingual Whisper choice. Better suited to capable local hardware than CPU-only systems.
Key facts
| Provider | OpenAI |
|---|---|
| Model family | Whisper |
| Parameters | 769M |
| License | MIT · Permissive — license text ↗ |
| Language coverage | 99+ listed languages · 100 enumerated in our record |
| Streaming | Streaming via a runtime or chunking wrapper |
| Timestamps | Yes — in the model or its runtime family |
| Speaker diarization | No — needs a separate component |
| Speech translation | Yes |
| Documented runtimes | openai-whisper, Hugging Face Transformers, whisper.cpp, faster-whisper — how to run it |
| Primary source | github.com/openai/whisper ↗ |
| Last reviewed |
Hardware and workload fit
These are Speech Model Match's conservative 1–5 editorial ratings, relative to the other 19 reviewed models. They come from model size, official memory guidance and documented runtimes — not from a universal benchmark. How the ratings work.
| Dimension | Rating | Scale |
|---|---|---|
| CPU-only fit | 2/5 | |
| Apple Silicon fit | 4/5 | |
| NVIDIA GPU fit | 5/5 | |
| Speed | 2/5 | |
| Accuracy (relative) | 5/5 | |
| Low memory use | 2/5 | |
| Long recordings | 5/5 | |
| Live / low latency | 2/5 | |
| Subtitle work | 5/5 | |
| Edge / low-power devices | 1/5 |
Where it is strong
- Apple Silicon fit: rated 4/5
- NVIDIA GPU fit: rated 5/5
- Accuracy (relative): rated 5/5
- Long recordings: rated 5/5
- Subtitle work: rated 5/5
- Speech translation is supported in the model family
- Timestamps are available in the model or its runtime family
What to check first
- CPU-only fit: rated 2/5
- Speed: rated 2/5
- Low memory use: rated 2/5
- Live / low latency: rated 2/5
- Edge / low-power devices: rated 1/5
- No native speaker diarization — pair it with a separate diarization pipeline if you need speaker labels
How it ranks in common setups
Ranks come from the same deterministic matcher that runs on the homepage, applied to all 26 reviewed models. They are decision guidance, not accuracy claims.
| Setup | Rank | Fit score | Top match |
|---|---|---|---|
| Long recordings on an 8–12 GB NVIDIA GPU, balanced | #5 of 26 | 76/100 · Strong fit | Whisper turbo |
| Live captions or voice input (native streaming + timestamps, speed first) | #16 of 26 | 70/100 · Possible fit | Nemotron 3.5 ASR Streaming 0.6B |
| Subtitles with timing, accuracy first | #1 of 26 | 87/100 · Excellent fit | This model |
| Meetings with speaker labels (diarization + timestamps) | #2 of 26 | 84/100 · Strong fit | VibeVoice-ASR 7B |
| CPU-only machine, low memory use | #14 of 26 | 59/100 · Possible fit | VibeVoice-ASR-BitNet |
| Apple Silicon with 8–16 GB unified memory, long recordings | #14 of 26 | 66/100 · Possible fit | Whisper small |
| Japanese live transcription (native streaming, speed first) | #12 of 26 | 64/100 · Possible fit | Nemotron 3.5 ASR Streaming 0.6B |
Supported languages
- Afrikaans
af - Albanian
sq - Amharic
am - Arabic
ar - Armenian
hy - Assamese
as - Azerbaijani
az - Bangla
bn - Bashkir
ba - Basque
eu - Belarusian
be - Bosnian
bs - Breton
br - Bulgarian
bg - Burmese
my - Cantonese
yue - Catalan
ca - Croatian
hr - Czech
cs - Danish
da - Dutch
nl - English
en - Estonian
et - Faroese
fo - Filipino
fil - Finnish
fi - French
fr - Galician
gl - Georgian
ka - German
de - Greek
el - Gujarati
gu - Haitian Creole
ht - Hausa
ha - Hawaiian
haw - Hebrew
he - Hindi
hi - Hungarian
hu - Icelandic
is - Indonesian
id - Italian
it - Japanese
ja - Javanese
jv - Kannada
kn - Kazakh
kk - Khmer
km - Korean
ko - Lao
lo - Latin
la - Latvian
lv - Lingala
ln - Lithuanian
lt - Luxembourgish
lb - Macedonian
mk - Malagasy
mg - Malay
ms - Malayalam
ml - Maltese
mt - Mandarin Chinese
zh - Māori
mi - Marathi
mr - Mongolian
mn - Nepali
ne - Norwegian
no - Norwegian Nynorsk
nn - Occitan
oc - Pashto
ps - Persian
fa - Polish
pl - Portuguese
pt - Punjabi
pa - Romanian
ro - Russian
ru - Sanskrit
sa - Serbian
sr - Shona
sn - Sindhi
sd - Sinhala
si - Slovak
sk - Slovenian
sl - Somali
so - Spanish
es - Sundanese
su - Swahili
sw - Swedish
sv - Tajik
tg - Tamil
ta - Tatar
tt - Telugu
te - Thai
th - Tibetan
bo - Turkish
tr - Turkmen
tk - Ukrainian
uk - Urdu
ur - Uzbek
uz - Vietnamese
vi - Welsh
cy - Yiddish
yi - Yoruba
yo
Compare coverage across every reviewed model in the language support matrix.
Community activity on Hugging Face
- Downloads, last 30 days
- 359,082
- Likes
- 317
- Repository last updated
- 29 Feb 2024
- Published on the Hub
- 26 Sept 2022
Figures from the Hugging Face Hub API, 11 Nov 2026. Source: huggingface.co/openai/whisper-medium ↗. Downloads count file downloads, not active users, and do not affect the matcher's scores. Compare all models.
How to run Whisper medium locally
OpenAI's reference package runs Whisper on PyTorch; faster runtimes such as whisper.cpp and faster-whisper load the same weights. Runtimes below are the ones the official model card or repository documents for this model (checked November 2, 2026).
- openai-whisper ↗Official package · PyTorch reference; needs ffmpeg on the system
- Hugging Face Transformers ↗Framework · pipeline API on GPU or CPU
- whisper.cpp ↗Third-party runtime · C/C++; CPU, Apple Metal/Core ML, CUDA, Vulkan
- faster-whisper ↗Third-party runtime · CTranslate2; INT8 on CPU, no system ffmpeg needed
Install command shown in the official github.com page ↗:
pip install -U openai-whisper
Audio input: openai-whisper decodes files with ffmpeg and processes audio in a sliding 30-second window. The whisper.cpp command-line tool expects 16-bit WAV; its README converts with ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav.
Official hardware guidance: OpenAI lists about 5 GB of VRAM for medium (769M parameters); whisper.cpp lists about 2.1 GB of memory. Source ↗
Whatever runtime you choose, measure speed and accuracy on your own audio with the benchmark tool; results differ a lot between runtimes, precision and hardware.
Before you deploy Whisper medium
A printable checklist built from the facts on this page. Tick it off before putting the model into production.
- Language. Confirm your language is in the 100-language list on this page, and test any accents or code-switching you expect.
- License. MIT: Include the copyright notice and the permission notice in all copies or substantial portions. Provided as is, without warranty. License guide.
- Install. Start from the official command:
pip install -U openai-whisper(full run notes). - Audio input. openai-whisper decodes files with ffmpeg and processes audio in a sliding 30-second window. The whisper.cpp command-line tool expects 16-bit WAV; its README converts with ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav.
- Hardware. OpenAI lists about 5 GB of VRAM for medium (769M parameters); whisper.cpp lists about 2.1 GB of memory.
- Features you rely on. For Whisper medium, streaming only through a runtime or chunking wrapper; timestamps are documented; speaker labels need a separate diarization component. Test each one you need with real audio.
- Measure. Run 10–20 representative clips, score them with the WER calculator and record speed in the benchmark workspace before you commit.
Alternatives and related models
Same family: Whisper tiny, Whisper base, Whisper small, Whisper turbo.
Similar fit profile: Qwen3-ASR 1.7B (Qwen, 1.7B), Distil-Whisper large-v3 (Hugging Face / Distil-Whisper, large-v3 distilled), Parakeet TDT 0.6B v3 (NVIDIA, 600M). Similarity is computed from the ten fit ratings plus streaming and diarization support.
| Fact | Whisper medium | Whisper tiny | Whisper base |
|---|---|---|---|
| Provider · size | OpenAI · 769M | OpenAI · 39M | OpenAI · 74M |
| License | MIT Permissive | MIT Permissive | MIT Permissive |
| Languages | 99+ listed languages | 99+ listed languages | 99+ listed languages |
| Streaming | Via runtime | Via runtime | Via runtime |
| Timestamps | Yes | Yes | Yes |
| Speaker diarization | Separate pipeline | Separate pipeline | Separate pipeline |
| CPU · Apple · NVIDIA fit | 2/5 · 4/5 · 5/5 | 5/5 · 5/5 · 4/5 | 5/5 · 5/5 · 4/5 |
| Live · long recordings | 2/5 · 5/5 | 3/5 · 3/5 | 3/5 · 3/5 |
| Main runtime | openai-whisper | openai-whisper | openai-whisper |
Head-to-head pages: Whisper medium vs Whisper turbo
Compare Whisper medium side by side
Guides that cover Whisper medium
- Moonshine vs Whisper: live voice or general-purpose ecosystem?
- Omnilingual ASR vs Whisper: breadth or ecosystem?
- Qwen3-ASR vs Whisper: which local ASR family fits better?
- Best local Whisper alternatives in 2026
- Whisper vs faster-whisper: same model family, different runtime
- Whisper vs Parakeet: which local ASR should you test first?
Check Whisper medium for your setup
Pick your language, hardware and workload to see this model's fit score, the reasons behind it and where it ranks against all reviewed models. Everything runs in your browser.
Common questions
Can I use Whisper medium commercially?
Permissive license; commercial use is allowed under the listed terms. The listed license is MIT; read the current license text ↗ before shipping a product, because terms can change. The license guide compares the obligations of all 26 reviewed models.
Does Whisper medium support real-time streaming?
Not natively. Live use relies on a runtime or chunking wrapper around a batch model, which adds latency and can split words at chunk borders. If live latency is critical, compare it with a native streaming model such as Nemotron 3.5 ASR Streaming 0.6B.
Which languages does Whisper medium transcribe?
99+ listed languages. The full list of 100 explicitly enumerated languages is shown above.
Can Whisper medium tell speakers apart?
Not natively. Combine it with a dedicated diarization component and evaluate that component separately; the meetings guide covers this workflow.
Primary sources
Reviewed against primary sources on September 16, 2026. Model behavior, runtime support, language coverage and licenses can change; re-check the linked primary source before production use. See the full source directory.