whatmodel.sh / stt / SeamlessM4T v2 Large
SeamlessM4T v2 Large
↗ share
+ contribute
SeamlessM4T v2 Large is an open-weights speech-to-text model from Meta, released November 2023. Its published size is 2.31B parameters. Licence: cc-by-nc-4.0.
Catalogue identity notes
Exact checkpoint facebook/seamless-m4t-v2-large (SeamlessM4T v2, the multitask speech/text translation model used in its ASR mode, target language = source language). Distinct from SeamlessM4T Large v1 (facebook/seamless-m4t-large) and the Medium model. Weights CC-BY-NC-4.0. Params: HF safetensors total 2,309,249,669 (card: 2.3B). released: HF repo created 2023-11-29 with the Seamless release. No local command added: the card documents only Python API usage.
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 SeamlessM4T v2 Large, out of 176 with a score, then more from Meta.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 84 | Omnilingual ASR CTC 7B | 55±2 | local | ||
| 85 |  Parakeet CTC 1.1B | 55±3 | local | ||
| 86 | SeamlessM4T v2 Large | 55±4 | local | ||
| 87 |  Qwen3-ASR-0.6B | 55±1 | local | ||
| 88 |  Ink-Whisper | 55±3 | api price unknown | ||
| more from Meta | |||||
| 1 | Muse Voice Transcribe 1.0 | 71±3 | api $0.0030/audio min | ||
| 58 | Omnilingual ASR LLM 7B | 58±2 | local | ||
| 59 | Omnilingual ASR LLM 7B v2 | 58±3 | local | ||
see all speech-to-text models · find the ones that fit your machine
$ what it's good at
multilingual transcription▂··▃▄█▆▅▄▂·▃ #2 of 36 * · 7.0% of its score · 1 test
Beats Universal-3.5 Pro, Cohere Transcribe Arabic (07-2026), Cohere Transcribe (03-2026) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Iberian ASR benchmark · 7-language macro WER (eu, ca, gl, pt, es + de, tr) | 8.1% ✎ | █▃▃··▅▃▃··▃▃ beats 9/10 | 7.0% |
other tests7 not in a task · 88% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Multilingual LibriSpeech 6 languages, mean WER (NVIDIA set) | 5.4% ✎ | █·▄········▄ beats 3/3 | 24% |
| Danish ASR Leaderboard · mean WER over 5 Danish test sets (RyeAI) | 32.2% ✎ | ▂▂···▃▂▅█▃▂▂ beats 5/16 | 16% |
| Common Voice English WER (version unspecified) | 16.2% ✎ | █···█·█····█ beats 3/3 | 15% |
| Open Universal Arabic ASR Leaderboard · 6-test-set average WER | 38.2% ✎ | ▃·▂▂▅█▅▃▂▅▂▂ beats 10/22 | 12% |
| TurkASR-Bench (7 Turkish sources, 5,200 utterances) · corpus WER | 21.6% ✎ | ▃··▃▃··▅··█▅ beats 4/9 | 10% |
| Quranic ASR Leaderboard · held-out recitation (600 clips) overall WER | 25.9% ✎ | ▃·▅·▅▃▃·▃▃▃█ beats 3/12 | 7.4% |
| FLEURS English test WER | 5.8% ✎ | ▁··▁▁▁▁▂▃█▂▁ beats 21/64 | 4.8% |
$ where to run it
Other ways to run it
- Run it as published · needs ~9.24 GB · from github.com/facebookresearch (official)
git clone https://github.com/facebookresearch/seamless_communication && cd seamless_communication && pip install . && m4t_predict audio.wav --task asr --tgt_lang eng --model_name seamlessM4T_v2_large
Weights and documentation: huggingface.co/facebook ↗
$ how the score was worked out · Canary-1B-v2 & Parakeet-TDT-0… 24%, ai-sage/GigaAM-Multilingual m… 20%, Danish Speech-to-Text Leaderb… 16% and 4 more
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
$ improve this page
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