Models / Qwen/ Qwen3 ASR 1.7B

Qwen3 ASR 1.7B

Qwen · Qwen/Qwen3-ASR-1.7B

Speech to textTranscribes a recording into words

Input: audio. Output: text.InputOutput
Type
Open weightsApache License 2.0
Languages
52
Size
2B

Context measured in tokens

Our take

Written Aug 2, 2026

Qwen3 ASR is a tiny downloadable speech-to-text model from Qwen that turns audio into written words. It runs extremely fast on benchmark hardware and carries a permissive Apache licence, though its accuracy on challenging audio drops sharply and no hosted offers are currently available.

Who should pick it

Pick this for offline or self-hosted transcription where licence permissiveness matters — Apache 2.0 allows commercial use and modification. Use it for batch processing of large audio archives, or for multilingual deployment needing broad language coverage. Skip it if you need verified accuracy beyond English, hosted availability without self-hosting, or reliable transcription of accented speech or recorded meetings.

The case for it

  • Extremely fast transcription for batch workloads: 394 times real time, processing roughly an hour of audio in 9 seconds on benchmark hardware.
  • Strong performance on clean read-aloud audio, with a word error rate of 1.24% versus 5.0% overall on the Open ASR benchmark.
  • Permissive open licence with no usage restrictions: Apache License 2.0 permits commercial use, redistribution and modification.
  • Broad language coverage relative to model size: 52 languages supported at 2.04 billion parameters.

The case against it

  • Accuracy collapses on challenging real-world audio: 9.88% word error rate on accented speech, nearly eight times worse than on clean read speech; 8.31% on recorded meetings.
  • No verified pricing or hosted availability: zero offers currently in our catalogue, with no release date disclosed.
  • English-only accuracy verification despite multilingual claims: all accuracy figures we hold are English-only, with no source measuring the other 51 languages.
00

How good is it?

TranscriptionTurning speech into text4 of 5Open ASR WER · 15th of 74

Words it gets right

95%

Misses roughly one word in 20, averaged over nine English test sets.

How fast it listens

394×43rd of 62

an hour of audio in 9 seconds, on the board's own hardware. Your machine will differ.

Languages

52

Stated by the leaderboard; we do not hold the list itself.

Where it struggles
Read aloudaudiobooks, clean recording1.2%
Podcasts and videoeveryday internet audio7.2%
Accented speechspeakers from many countries9.9%
Meetingsa room, several people, far microphone8.3%

Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case, not to the board.

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.

Also scored, on boards we give no mark for
Podcasts and video 5th of 74Recorded meetings 20th of 74European-accented speech 20th of 74Accented speech 23rd of 74Clean read speech 25th of 74Harder read speech 29th of 74Financial calls 33rd of 74

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.
394.1independentsource ↗
5independentsource ↗
9.9independentsource ↗
2.6independentsource ↗
8.3independentsource ↗
7.2independentsource ↗
1.2independentsource ↗
2.9independentsource ↗
01

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

A card many people ownFits in memory

GeForce RTX 4090 · 24 GB

Weights at Q4_K_M1.3 / 24 GBest
Spare memory20.2 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed652 tok/sest

Room to spare. 20.2 GB spare means a 10% error in the size would not change the answer.

One step upFits in memory

GeForce RTX 5090 · 32 GB

Weights at Q4_K_M1.3 / 32 GBest
Spare memory28.2 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed1159 tok/sest

Room to spare. 28.2 GB spare means a 10% error in the size would not change the answer.

On a MacFits in memory

Apple M2 (8-core GPU, 8GB unified) · 8 GB

Weights at Q4_K_M1.3 / 8 GBest
Spare memory3.4 GB spare
Usable context131Kwhat the spare memory holds; no published limit on record
Decode speed57 tok/sest

Room to spare. 3.4 GB spare means a 10% error in the size would not change the answer.

Your hardware
Checking your profile…

Memory use by level

Against a 24 GB card.

Q4_K_M
recommended
1.3 GBest
Fits in memory
Q5_K_M
1.5 GBest
Fits in memory
Q8_0
2.3 GBest
Fits in memory

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 →

02

Models people weigh against Qwen3 ASR 1.7B

03

When we formed this view

Dates behind this page

Aug 2, 2026BenchmarkScored 9.9 on Accented speechleaderboard
Aug 2, 2026BenchmarkScored 2.6 on Financial callsleaderboard
Aug 2, 2026BenchmarkScored 8.3 on Recorded meetingsleaderboard
Aug 2, 2026BenchmarkScored 3 on European-accented speechleaderboard
Aug 2, 2026BenchmarkScored 7.2 on Podcasts and videoleaderboard
Aug 2, 2026BenchmarkScored 1.2 on Clean read speechleaderboard
Aug 2, 2026BenchmarkScored 2.9 on Harder read speechleaderboard
Aug 1, 2026ListedListed on LLMapfirst indexed by our pipeline
Aug 1, 2026BenchmarkScored 394.1 on Open ASR RTFxleaderboard
Aug 1, 2026BenchmarkScored 5 on Open ASR WERleaderboard

Prices last checked 7h 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.
04

Licence and identifiers

What the licence allowsApache License 2.0, 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

Apache License 2.0

permissiveCommercial use allowed

Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.

Identifiers

Modality record
audio->text
Catalogue slug
qwen-qwen3-asr-1-7b

Machine-readable model card (omc.json) →

Something wrong on this page? Tell us