MMS 1b All
Meta · released May 27, 2023 · facebook/mms-1b-all
Speech to textTranscribes a recording into words
- Type
- Open weightsCreative Commons Attribution-NonCommercial 4.0
- Languages
- 129
- Size
- 1B
Context measured in tokens
Our take
Written Aug 2, 2026MMS 1b All is a one-billion-parameter speech-to-text model from Meta that converts audio to written words across 129 languages. Its non-commercial licence and high error rate on challenging audio make it a research tool rather than a production default.
Pick this for research where extreme multilingual coverage matters — it handles low-resource languages such as Abkhaz, Afar and Aymara that most models skip. Use it for batch transcription where speed is key: an hour of audio in about two seconds. Skip it if you need commercial deployment, messy or accented audio, or strong real-world English accuracy.
The case for it
- Extremely fast batch transcription at 1,954 times real time — an hour of audio in roughly two seconds on benchmark hardware.
- Broadest language coverage in our speech catalogue, with 129 languages including low-resource languages such as Abkhaz, Afar, Akan, Amharic, Arabic, Assamese, Avar and Aymara.
- Strong on clean read-aloud English, with about three words wrong per hundred.
The case against it
- High overall English error rate of 13.47% on the Open ASR benchmark — roughly one word in seven is wrong.
- Accuracy collapses in challenging conditions: 31.24% on recorded meetings, nearly ten times worse than its clean-speech result; 23% on accented speech, more than seven times worse.
- Non-commercial licence blocks commercial use, and no hosted offers are tracked.
How good is it?
TranscriptionTurning speech into text1 of 5Open ASR WER · 74th of 74
86.5%
Misses roughly one word in 7, averaged over nine English test sets.
1,954×20th of 62
an hour of audio in 2 seconds, on the board's own hardware. Your machine will differ.
129
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, not to the board.
Which languages ↓ ↑
ab · Afrikaans · Akan · Amharic · Arabic · Assamese · av · Aymara · Azerbaijani · ba · Bambara · Belarusian · Bangla · bi · bo · Serbian (Latin) · Breton · Bulgarian · Catalan · Czech · ce · cv · Kurdish · Welsh · Danish · German · Divehi · dz · Greek · English · Esperanto · Estonian · Basque · Ewe · Faroese · Persian · fj · Finnish · French · Western Frisian · ff · Irish · Galician · Guarani · Gujarati · Chinese · Haitian Creole · Hausa · Hebrew · Hindi · Hungarian · Armenian · Igbo · Interlingua · Malay · Icelandic · Italian · Javanese · Japanese · Kannada · Georgian · Kazakh · kr · Khmer · ki · Kinyarwanda · Kyrgyz · Korean · kv · Lao · Latin · Latvian · Lingala · Lithuanian · Luxembourgish · Ganda · mh · Malayalam · Marathi · Macedonian · Malagasy · Maltese · Mongolian · Māori · Burmese · Dutch · Norwegian · Nepali · Nyanja · Occitan · Oromo · Odia · os · Punjabi · Polish · Portuguese · Pashto · Quechua · Romanian · rn · Russian · sg · Slovak · Slovenian · Samoan · Shona · Sindhi · Somali · Spanish · Albanian · Sundanese · Swedish · Swahili · Tamil · Tatar · Telugu · Tajik · Filipino · Thai · Tigrinya · Tsonga · Turkish · Ukrainian · Vietnamese · Wolof · Xhosa · Yoruba · Zulu · za
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.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
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?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 20.9 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. 22.1 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. 4.1 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
When we formed this view
Dates behind this page
Prices last checked 6h ago
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 allowsCreative Commons Attribution-NonCommercial 4.0, 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
Creative Commons Attribution-NonCommercial 4.0
Weights are downloadable but commercial use is prohibited. Research and personal use only.
Identifiers
- Hugging Face
- facebook/mms-1b-all
- Architecture
- Dense
- Modality record
- audio->text
- Catalogue slug
- facebook-mms-1b-all