- Status
- Live
- Languages
- 30 languages
- License
- Apache 2.0
Model ID
| Model ID | Designed for |
|---|---|
qwen3-asr | Multilingual transcription with Qwen3-ASR 1.7B |
Confirm access with GET /v1/models, then follow the utility-model docs for the canonical v1 REST or native gRPC request contract.
Pricing
qwen3-asr: Free for a limited time. $0.00003 $0.00 per audio second
Overview
Tarka’s qwen3-asr route uses Qwen/Qwen3-ASR-1.7B, an
Apache 2.0 transcription option. Use the qwen3-asr alias with
POST /v1/audio/transcriptions or the native gRPC transcription
method.
This route uses Qwen3-ASR, not Qwen-Audio-3.1. The upstream release documents
30 languages and 22 Chinese dialects; dialects are not additional languages.
Nepali is not in that supported-language list. For Nepali, start with
whisper-nepali-medium and evaluate your own recordings.
curl --fail-with-body https://tarka.rest/v1/audio/transcriptions \
-H "Authorization: Bearer $TARKA_API_KEY" \
-F "model=qwen3-asr" \
-F "file=@speech.wav" \
-F "language=en" \
-F "response_format=json"When to pick Qwen3-ASR over Whisper
- License hygiene matters. Whisper is MIT — close to Apache 2.0 — but some procurement teams prefer Apache 2.0 with its explicit patent grant and familiar enterprise terms. Qwen3-ASR is the Apache option.
- Your language is supported. Compare transcripts on your own audio; model size or language coverage does not establish better accuracy.
When to pick Whisper instead
- Broader language coverage. Whisper covers more languages. Verify that the selected checkpoint supports the language you need.
- Mature community tooling. Whisper’s ecosystem (timestamps, diarization, translation-to-English, fine-tuning recipes) is more mature.
- Reference benchmark numbers. Most academic and product benchmarks reference Whisper; staying on the same model simplifies comparison.
Qwen3-ASR provides transcription, not translation-to-English, on the current
Tarka route. Use whisper-large-v3 with
/audio/translations when translation is required. Use model
discovery and your organization limits as the source of truth for access and
request limits.
The current Tarka route returns a completed transcription; it does not expose live streaming or speaker diarization. Upstream forced-alignment support does not enable word timestamps in this deployment: timestamp requests are rejected. See the utility API guide for the shared request fields and service status for current health.