Qwen-Audio-3.1-Realtime
Full-duplex speech model for voice agents that decide when to call tools
Paid
Audio Generation
Undisclosed
parameters
Proprietary
license
Sep 2026
released
Alibaba released Qwen-Audio-3.1-Realtime on September 28, 2026 as the flagship of a five-model Qwen-Audio-3.1 stack that also covers speech recognition and text-to-speech. It’s built for full-duplex voice agents, meaning it can listen and respond at the same time and decide mid-conversation when to speak versus when to call an external tool.
// similar models in Audio Generation
Eleven v4
ElevenLabs' top-ranked text-to-speech model with a real-time Turbo variant
ElevenLabs · Audio Generation · Paid
Whisper
OpenAI's open-weight speech recognition and transcription model
OpenAI · Audio Generation · Free
Gemini 3.8 Live
Google's real-time speech-to-speech model for production voice agents
Google DeepMind · Audio Generation · Paid
GPT-Live-1
OpenAI's full-duplex voice model that listens and speaks at the same time
OpenAI · Audio Generation · Freemium
Eleven Multilingual v2
ElevenLabs' flagship text-to-speech model across 29 languages
ElevenLabs · Audio Generation · Freemium
Eleven v3
ElevenLabs' most expressive text-to-speech model with fine-grained emotional control
ElevenLabs · Audio Generation · Freemium