← All models

Qwen-Audio-3.0-Realtime-Plus

Published Publisher Alibabaqwen-audio-3.0-realtime-plus

Positioning and release

Qwen-Audio-3.0-Realtime-Plus is Alibaba’s standard-edition real-time duplex speech dialogue model, offered through Alibaba Cloud Model Studio. It emphasizes response quality and natural dialogue. The official China release listing dates it July 14, 2026; the same listing records international deployment on August 10, 2026. (help.aliyun.com)

Capabilities and uses

  • Accepts audio and text input and produces audio and text output. It supports function calling and web search. The speech-to-speech guide says web search is enabled with enable_search, but cannot be used simultaneously with function calling for this model. (alibabacloud.com)
  • Designed for real-time speech conversations; potential applications include voice assistants and interactive customer service.

Limits

  • Context window: 40,960 tokens; maximum input: 16,384 tokens; maximum output: 8,192 tokens. (alibabacloud.com)
  • Structured outputs, prefix completion, context caching, batch inference and fine-tuning are not supported. The listed rate limits for Beijing and Singapore are 60 requests per minute and 100,000 tokens per minute. (alibabacloud.com)

Official API pricing

Original prices, excluding time-limited promotions; checked November 16, 2026. All amounts below are per 1 million tokens. These are separate platform and region listings, not currency conversions. (alibabacloud.com)

Official platform and deployment region Currency Input text Input audio Text-only output Audio output with accompanying text
Alibaba Cloud China, Beijing CNY 5 40 40 150
Alibaba Cloud China, Singapore CNY 5.995 47.963 47.963 179.861
Alibaba Cloud international, Beijing USD 0.688 5.501 5.501 20.628
Alibaba Cloud international, Singapore USD 0.8 6.4 6.4 24

When text and audio are output together, the accompanying output text is not separately charged. The China pricing guide states that audio is billed at 12.5 tokens per second, with durations under one second rounded up to one second. Conversation history can be billed again as input in later turns. (alibabacloud.com)

Official sources