Cover: AI-generated editorial illustration.
Evergreen guide. Prices and details checked on 13 November 2026 against the linked pages; prices can change, so check them before you buy.
Qwen3 ASR is available on Opennan as two models: qwen3-asr-0.6b at $0.00026 per minute and qwen3-asr-1.7b at $0.000585 per minute. Both are billed per started minute and use the same API key as every other Opennan model.
Model IDs and what they accept
The two callable model IDs are qwen3-asr-0.6b and qwen3-asr-1.7b. Both accept one required input: audio_url, a public HTTPS link to an audio or video file. The maximum input length documented on each model page is 30 minutes per file. Both models detect the spoken language automatically. You can also pass an optional language parameter as a 2- or 3-letter code, for example en, th, or ja, to specify the language directly rather than letting the model detect it.
Sending a request and reading the result
Both models share one endpoint: POST https://api.opennan.com/v1/audio/transcriptions. A call returns a task ID immediately. You then poll GET https://api.opennan.com/v1/tasks/{id} until the task status becomes succeeded or failed.
Request for Qwen3-ASR 0.6B
curl https://api.opennan.com/v1/audio/transcriptions -H "Authorization: Bearer $OPENNAN_KEY" -H "Content-Type: application/json" -d '{"model":"qwen3-asr-0.6b","audio_url":"https://example.com/speech.mp3"}'
Request for Qwen3-ASR 1.7B
curl https://api.opennan.com/v1/audio/transcriptions -H "Authorization: Bearer $OPENNAN_KEY" -H "Content-Type: application/json" -d '{"model":"qwen3-asr-1.7b","audio_url":"https://example.com/speech.mp3"}'
The task response follows the shape the Opennan API docs describe for all asynchronous tasks:
{"id":"task_…","status":"succeeded","progress":100,"outputs":[…],"price_usd":0.00026,"error":null}
Statuses move from queued to running and then to succeeded or failed. When a task fails, the error object contains "refunded": true and you are not charged.
Choosing between 0.6B and 1.7B
Both models share the same endpoint, the same billing unit, and the same 30-minute input limit. The difference is price. The 0.6B model costs $0.00026 per minute, which the 0.6B model page shows is 87% below Alibaba's official list price of $0.0021 per minute. The 1.7B model costs $0.000585 per minute, which the 1.7B model page shows is 72% below that same official rate. The 0.6B is the lower-cost option; the 1.7B is a larger model at a higher price point. Both are priced significantly below the maker's published rate.
One practical advantage of calling either model on Opennan is that the same key and balance also cover every other model on the platform, video, images, text, music, 3D, and the other transcription models, without separate accounts or separate top-ups.
Price comparison
All prices are per minute of audio, billed per started minute. Opennan prices were checked October 26, 2026, as shown on the Opennan pricing page and each model page. The official prices are the model maker's own list prices as shown on those same pages.
| Provider and model | Price per minute | Notes |
|---|---|---|
| Opennan, Qwen3-ASR 0.6B | $0.00026 | 87% below Alibaba official |
| Opennan, Qwen3-ASR 1.7B | $0.000585 | 72% below Alibaba official |
| Opennan, Whisper Large v3 Turbo | $0.00026 | 95% below OpenAI official |
| Opennan, Whisper Large v3 | $0.000585 | 90% below OpenAI official |
| Alibaba (Qwen) official, Qwen3-ASR | $0.0021 | Maker's list price, applies to both tiers |
| OpenAI official, Whisper | $0.006 | Maker's list price, applies to both v3 tiers |
Switching between models requires only a change to the model field. The endpoint, key, and balance stay the same.
FAQ
Does Qwen3 ASR return word-level timestamps?
The Opennan model pages and docs do not list word timestamps as a parameter or output field for qwen3-asr-0.6b or qwen3-asr-1.7b. The documented output is transcript text and a detected language code.
Does Qwen3 ASR support speaker diarization?
The model pages do not mention diarization for either model. The parameter tables on each page list only model and audio_url as accepted fields, with language as an optional addition on the shared speech-to-text endpoint.
Which languages does Qwen3 ASR recognize?
Both models detect the spoken language on their own. You can override this by passing the language parameter with a 2- or 3-letter code such as en, th, or ja. The model pages do not publish a fixed list of supported language codes.
Is live streaming transcription available?
POST /v1/audio/transcriptions is a task-based endpoint. The Opennan API docs list a separate WSS /v1/realtime/stt endpoint for streaming live transcription, but that is a different endpoint from the two Qwen3-ASR models described here.
Am I charged if a transcription run fails?
No. Both model pages and the API docs state that failed tasks are refunded in full. You pay only for completed results.
Get started with Qwen3 ASR on Opennan
Both models are live now. You can try Qwen3-ASR 0.6B and Qwen3-ASR 1.7B directly in the browser on each model page, or visit the speech-to-text API page to compare all six transcription models on the same endpoint. One API key covers both Qwen3-ASR models and every other Opennan model in the same balance.