Qwen: Qwen3 ASR 0.6B

Qwen3 ASR 0.6B is a compact automatic speech recognition model from Qwen. It belongs to the transcription category, so its job is turning speech into written text rather than generating conversational replies or handling general purpose language tasks. The 0.6B in the name indicates a small parameter count, which makes it a lightweight option for speech recognition work compared with larger models.

The model accepts audio as its input and produces a transcription as its output. It supports multilingual language identification and transcription across 30 languages and 22 Chinese dialects, so it can both detect which language is being spoken and transcribe the audio in that language. It handles streaming and offline transcription. No context length is published for this model, and its maximum output is not defined in the model metadata, so you should treat it primarily as a speech to text component rather than something you feed long text prompts into.

Because of those capabilities, the natural use cases are speech to text pipelines, meeting or call transcription, subtitle and caption generation for audio and video, voice note transcription, and multilingual applications where you need language identification before or alongside transcription. The Chinese dialect coverage makes it relevant for products serving Chinese speaking users whose audio does not match standard Mandarin. Streaming support suits live or near real time transcription, while offline mode fits batch processing of recorded files.

You call the model through api14's OpenAI-compatible API using the model id "qwen/qwen3-asr-0.6b". Because the endpoint follows the OpenAI format, you can point an existing OpenAI client at api14 and set that model id, sending audio as the input. No supported request parameters are listed for this model, so you should not assume tuning options such as temperature or top_p are available, and the request shape is essentially the audio input plus the model id. Check the live price table on this page for current rates.

One practical limit to keep in mind is scope. This is a transcription model with audio input and transcription output, not a chat or reasoning model, so tasks like summarising a transcript, answering questions about it, or generating text from instructions belong to a different model. If you need those follow up steps, run the transcription here first and pass the result to a text model.