whatmodel.sh / stt / Qwen3-ASR-1.7B

Qwen3-ASR-1.7B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen3-ASR-1.7B is an open-weights speech-to-text model from Alibaba (Qwen), released January 2026. Its published size is 2.35B parameters. Licence: apache-2.0.

60.1±1.2#32 of 162 speech-to-text modelsfrom 33 benchmark results across 16 sources

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 Qwen3-ASR-1.7B, out of 176 with a score, then more from Alibaba (Qwen).

#modelscorewhereRAMreleased
37Parakeet Unified EN 0.6B NVIDIA60±4local~0.7GB2026-04
38Grok Voice Transcribe 1.0 xAI60±5comparison only2026-04
39Qwen3-ASR-1.7B Alibaba (Qwen)60±1api $0.0005/audio min +local~3.3GB2026-01
40Canary-1B-Flash NVIDIA60±2local~4.4GB2025-03
41Granite Speech 3.3 2B IBM60±3local~7.7GB2025-06
more from Alibaba (Qwen)
87Qwen3-ASR-0.6B Alibaba (Qwen)55±1local~2.8GB2026-01

see all speech-to-text models · find the ones that fit your machine

$ what it's good at

multilingual transcription▂··▃▄█▆▅▄▂·▂ #21 of 36 · 22% of its score · 3 tests

Trails Omnilingual ASR CTC 1B v2, Omnilingual ASR CTC 7B v2, SeamlessM4T v2 Large here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Open ASR multilingual WER5.1% ✎▂··▂▃▂█▃█▂▁▁ beats 17/29avg 8.9%▇▇▇▇▇▇▇ 18%typical
QCRI Arabic-English ASR leaderboard, 22-set average WER29.3% ✎█···█······█ beats 1/2avg 53%▇ 3.0%small run
Iberian ASR benchmark · 7-language macro WER (eu, ca, gl, pt, es + de, tr)39.9% ✎█▃▃··▅▃▃··▃▃ beats 4/10avg 38%▇ 0.9%noisy
other tests27 not in a task · 78% of its score
testresultvs the others in its rundifficultycountswhy
LibriSpeech test-clean WER +11.6% ✎▁···▁▁▂·▄▄██ beats 43/56avg 2.3%▇▇▇ 7.6%blunt · several runs
Mega-ASR adverse-condition suite · mean WER (CHiME-4, VOiCES, NOIZEUS)7.9% ✎▄····█▄▄···▄ beats 5/5avg 19%▇▇ 5.9%typical
Open ASR English short-form WER (March 2026)5.8% ✎▁▁▁▃··▁▁▂██▆ beats 81/86avg 8.2%▇▇ 5.3%blunt
RW-Voice-EQ ASR · 4 real-world English sets, mean WER7.4% ✎█▃▃▅▃···▃··▃ beats 8/9avg 15%▇▇ 4.2%typical
AfriCodeSwitch English-Yoruba code-switched speech · WER (cap2000)67.6% ✎█··█·█···█·█ beats 3/4avg 131%▇▇ 3.9%typical
Dutch ASR Leaderboard · average WER (FLEURS, VoxPopuli, MLS nl)10.9% ✎▄··▄█···█··▄ beats 1/6avg 16%▇▇ 3.8%typical
FLEURS Mandarin test error rate +17.1% ✎▂▂··▂▄▂▂·▆██ beats 17/17avg 12%▇▇ 3.8%several runs
Multilingual LibriSpeech English test WER lab4.6% ✎█······█·█·█ beats 3/3avg 7.7%▇ 3.4%typical
19 more tests
testresultvs the others in its rundifficultycountswhy
FLEURS English test WER +13.2% ✎▁··▁▁▁▁▁▁▃█▃ beats 56/56avg 5.8%▇ 3.3%blunt · several runs
WenetSpeech test_net error rate lab5.0% ✎█···█·····██ beats 3/3avg 13%▇ 3.2%typical
AISHELL-2 test error rate lab2.7% ✎█··█·····█·█ beats 3/3avg 5.7%▇ 3.1%typical
GigaSpeechBench · 7 Arabic dialect WER mean48.3% ✎█··█·██··█·█ beats 2/5avg 79%▇ 3.1%typical
LibriSpeech test-other WER lab3.4% ✎▁▁▁▂▅▃▅▄█▃▂▁ beats 50/54avg 5.5%▇ 2.8%typical
GigaSpeechBench · 12 English domains entity B-WER mean13.8% ✎▄·······▄·█▄ beats 4/4avg 28%▇ 2.6%typical
PESTE · Persian FLEURS (fa_ir) corpus CER8.9% ✎▄··▄·▄▄▄··██ beats 5/8avg 12%▇ 2.6%typical
GigaSpeechBench · 6 English accents WER mean15.1% ✎▄······▄··▄█ beats 4/4avg 36%▇ 2.2%typical
Common Voice English WER (version unspecified) lab7.4% ✎█··▄·······▄ beats 3/3avg 13%▇ 2.2%typical
Common Voice Mandarin error rate (version unspecified) lab5.3% ✎█·······██·█ beats 3/3avg 11%▇ 2.2%typical
GigaSpeech test WER lab8.5% ✎▁····▁··▁▄█▃ beats 17/17avg 17%▇ 2.1%typical
Danish ASR Leaderboard · mean WER over 5 Danish test sets (RyeAI)29.5% ✎▂▂···▃▂▅█▃▂▂ beats 7/16avg 32%▇ 2.0%blunt
WenetSpeech test_meeting error rate lab5.9% ✎█··█······██ beats 3/3avg 21%▇ 2.0%sharp · noisy
VoxPopuli English test WER lab9.2% ✎█▄▄··████▄██ beats 2/16avg 13%▇ 1.7%typical
Open Universal Arabic ASR Leaderboard · 6-test-set average WER33.4% ✎▃·▂▂▅█▅▃▂▅▂▂ beats 16/22avg 40%▇ 1.5%blunt
TurkASR-Bench (7 Turkish sources, 5,200 utterances) · corpus WER15.8% ✎▃··▃▃··▅··█▅ beats 6/9avg 17%▇ 1.3%blunt
Quranic ASR Leaderboard · held-out recitation (600 clips) overall WER33.7% ✎▃·▅·▅▃▃·▃▃▃█ beats 2/12avg 24%▇ 0.9%blunt
Speech Accessibility Project dysarthric speech · zero-shot WER (rated subset, N=11,218)13.4% ✎▂·▂··▂····▆█ beats 9/9avg 18%▇ 0.6%blunt
TED-LIUM test WER (short-form) lab4.5% ✎▁·▁▇█▃·▁▁▃█▄ beats 17/27avg 6.2%▇ 0.1%blunt

$ where to run it

Through an API

api · DeepInfra · $0.0004/audio min · deepinfra.com ↗ · price sourceUSD per input audio minute: DeepInfra lists Qwen/Qwen3-ASR-1.7B at $0.00045 / minute (page pricing data: input_length billing). Product: DeepInfra hosted inference. API model: Qwen/Qwen3-ASR-1.7B. Settings: standard serverless. Price basis: published list rate. Currency: USD. DeepInfra official model page, observed 2026-09-25.

Other ways to run it

version

  1. Run it as published · needs ~3.32 GB, Mac only · from github.com/Blaizzy
    uvx --from mlx-audio mlx_audio.stt.generate --model mlx-community/Qwen3-ASR-1.7B-8bit --audio audio.wav --output-path out --language English

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen/Qwen3-ASR-1.7B model card 25%, Open ASR Leaderboard 23%, transcribe.cpp model catalog… 12% and 13 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

sourceresultsshare of the scorelicence
Qwen/Qwen3-ASR-1.7B model card · lab-reported13▇▇▇▇▇▇▇▇ 25%Apache-2.0
Open ASR Leaderboard2▇▇▇▇▇▇▇ 23%Apache-2.0
transcribe.cpp model catalog · LibriSpeech test-clean WER of validated full-precision ports3▇▇▇▇ 12%MIT
GigaSpeechBench: A Real-World Multilingual Speech-to-Text Benchmark3▇▇ 8%CC BY 4.0
Mega-ASR (arXiv:2605.19833v1) · adverse-condition ASR baselines (CHiME-4, VOiCES, NOIZEUS)1▇▇ 6%CC BY 4.0
RW-Voice-EQ Bench (Hume AI)1▇ 4%CC BY 4.0
Switch-aware evaluation of ASR on English-Yoruba code-switched speech (arXiv:2609.11786v1)1▇ 4%CC BY 4.0
Dutch ASR Leaderboard (tvosch) · committed result files1▇ 4%Apache-2.0
QCRI Arabic + English ASR leaderboard (Hugging Face Space QCRI/AURA-Arabic-English-ASR-leaderboard)1▇ 3%CC BY 4.0
PESTE · Persian Speech to Text benchmark (fleurs-fa-ir-v1)1▇ 3%Apache-2.0
Danish Speech-to-Text Leaderboard (RyeAI) · results dataset1▇ 2%MIT
Open Universal Arabic ASR Leaderboard (Elm Research Center)1▇ 2%Apache-2.0
Turkish ASR Leaderboard (TurkASR-Bench, yagmurtuncer)1▇ 1%MIT
Quranic ASR Leaderboard (Quran-Lab, held-out recitation set)1▇ 1%Apache-2.0
Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 31▇ 1%CC BY 4.0
Dysarthric ASR with clinical context on the Speech Accessibility Project (arXiv:2605.02782v2)1▇ 1%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 60.1

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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