Alibaba releases Qwen-Audio-3.1 voice model for realtime interaction
Alibaba’s Qwen team has released Qwen-Audio-3.1, an audio stack covering automatic speech recognition, text-to-speech and realtime interaction. Its main component, Qwen-Audio-3.1-Realtime, is a full-duplex model designed for voice agents that can reason, use tools and manage when to speak. It is available as a managed QwenCloud API through WebSocket under the name qwen-audio-3.1-realtime-plus; no open weights were announced.
The service accepts and produces text and audio, supports a 262K-token context, function calling, web search, structured outputs, context caching and fine-tuning. Listed pricing is $6.4 per 1M audio input tokens and $24 per 1M output tokens for text and audio, while text input costs $0.8 per 1M tokens. The rates were checked on September 28, 2026, and audio tokenization differs among providers.
The system combines a decision model with a speech-to-text model and a context-aware voice renderer. Compared with version 3.0, Audio MultiChallenge increased from 47.12 to 52.21, the 14-language BBA average from 81.7% to 88.1%, and FLEURS WER fell from 9.01 to 3.98. On Full-Duplex-Bench v1.5, replies to background speech declined from 73.0% to 13.0%. In a τ-Voice half-duplex adaptation, task success rose from 78.4% to 82.0%.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.