Specification Parameters
Basic Information
- Brand
- 通义千问
- Model
- qwen3-livetranslate-flash
- Key Parameters
- 音频输入格式 mp3/wav 等,输出音频格式 wav,语速范围 0.5–2.0,temperature 取值 [0,2),top_p 取值 (0,1.0]
Comparison with Similar Products
| Specifications | 本品 · qwen3-livetranslate-flash | kimi-k3.1-pro | 火山方舟模型 API | qwen-audio-3.1-asr-flash | doubao-seed-2.1-turbo |
|---|---|---|---|---|---|
| Main controller | — | — | — | — | — |
| NPU | — | — | — | — | — |
| Memory / VRAM | — | — | — | — | — |
| Reference unit price | Request a Quote | Request a Quote | Request a Quote | Request a Quote | Request a Quote |
| Inventory | — | — | — | — | — |
Product introduction
Interface & Deployment: The model is invoked via an OpenAI-compatible interface. The base_url is https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1 for the Beijing region and https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1 for the Singapore region; {WorkspaceId} is the workspace ID. The original documentation explicitly states that "invocation via the DashScope interface is not supported" and that "the model only supports streaming output calls; it can only be set to true".
Input & Output: For audio input, type is input_audio (data supports a URL or Base64, with formats such as mp3, wav); for video input, type is video_url (public URL or Base64 Data URL). The output modalities parameter can be ["text"] or ["text","audio"]; when outputting audio, voice and format must be configured, and format only supports wav. The message array supports only one User Message.
Translation & Sampling Parameters: translation_options is required; target_lang is mandatory, while source_lang is optional (auto-detected if not set). To ensure translation accuracy, the documentation does not recommend modifying sampling parameters such as temperature, top_p, presence_penalty, top_k, and repetition_penalty; speech_rate can control output speech rate within 0.5–2.0. top_k and repetition_penalty are non-OpenAI-standard parameters and must be placed in extra_body in the Python SDK.
Versions & Metering: Supported models are qwen3-livetranslate-flash and qwen3-livetranslate-flash-2025-12-01. Token usage is calculated based on prompt_tokens and completion_tokens. Tokens for input audio and video are counted separately in prompt_tokens_details.audio_tokens; audio tokens in video files are also counted in this field, and input text tokens are fixed at 0.
Integration Notes: The API keys differ between the Beijing and Singapore regions; the default and maximum values of max_tokens are both the model's maximum output length, and output exceeding it will be truncated. Before selection, consult the official "Supported Voices," "Supported Languages," and "Model Selection" documentation to confirm specific capability boundaries.