Qwen-Audio 3.0

Qwen-Audio-3.0-Realtime is Alibaba's real-time voice interaction model, launched July 15, 2026 in two tiers: Plus, the more reasoning-capable version, and Flash, the faster one. It is built for voice agents that talk, listen, and act: a full-duplex decision sub-model predicts whether to keep listening, speak, stop, or resume, while a speech-to-text model writes response content and a context-aware renderer turns it into streaming speech.

Standout capabilities include millisecond-level response, instruction-free tool calling, conversational memory, and empathetic tone adjustment based on context. Independent evaluator Artificial Analysis ranked the flagship above OpenAI's realtime model on its Speech-to-Speech Index. It is an API-only hosted model on Alibaba Cloud Model Studio; no open weights were announced.