Alibaba releases Qwen-Audio-3.0-Realtime, where Voice AI proactively calls upon tools to get things done

According to Vision OneMillion monitoring, Alibaba has released the real-time speech model Qwen-Audio-3.0-Realtime. It can respond in milliseconds and actively invoke external tools such as maps, APIs, MCP, and knowledge bases during conversations.

Users do not have to explicitly say "invoke a tool". For example, if they first ask about a nearby Sichuan restaurant and then inquire about one with a high rating and close distance, the model will remember the results of the previous query and continue searching and answering.

The model can also listen while speaking, allowing users to interrupt at any time. In noisy environments or during multi-party conversations, it will attempt to identify the primary speaker to reduce interruptions from background noise.

The speech will also adjust its tone, pace, and emotion based on the context. The official release includes a Plus version for enhanced inference and a Flash version for quicker response times, both of which are now available on Alibaba Cloud Hundred Refinements.