Qwen3 ASR 1.7B
Qwen/Qwen3-ASR-1.7B
published Jan 2026 · updated Jan 2026
Qwen3 ASR 1.7B is an ASR model that supports language identification and speech recognition for 52 languages and dialects.
specs
| Task | Automatic Speech Recognition |
| Architecture | Qwen3-Omni-based transformer |
| Parameters | 1.7B |
| License | Apache 2.0 |
about this model
Capabilities
The model handles 30 languages and 22 Chinese dialects, including English accents from multiple regions. It processes speech, singing voice, and songs with background music. A single model supports both offline and streaming inference, as well as long audio transcription.
Performance
Qwen3-ASR-1.7B achieves state-of-the-art results among open-source ASR models and is competitive with the strongest proprietary commercial APIs. The smaller Qwen3-ASR-0.6B offers an accuracy-efficiency trade-off, achieving an average time-to-first-token of 92 ms and transcribing 2,000 seconds of speech in 1 second at a concurrency of 128.
Architecture


Forced Alignment
An optional companion, Qwen3-ForcedAligner-0.6B, predicts timestamps for arbitrary units within up to 5 minutes of speech across 11 languages, outperforming existing end-to-end forced-alignment models in accuracy.
License & Source
Released under Apache 2.0. Full technical details: arXiv:2601.21337.
best for
- ·Multilingual speech transcription for global applications
- ·Real-time streaming ASR for voice assistants
- ·Forced alignment for subtitle generation
FAQ
It supports 30 languages and 22 Chinese dialects, totaling 52 languages and dialects.
It achieves state-of-the-art results among open-source ASR models and is competitive with proprietary commercial APIs.
It is released under the Apache 2.0 license.
Send requests to the gigarouter OpenAI-compatible endpoint with your API key; the endpoint accepts audio URLs, file paths, or raw audio data.
Audio can be provided as a local path, URL, base64-encoded data, or a (numpy array, sample rate) tuple.
We're benchmarking and onboarding Qwen3 ASR 1.7B as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.