Alibaba Cloud Qwen-Audio-3.1-Realtime

Qwen-Audio-3.1-Realtime is Alibaba Cloud's native speech-to-speech model released in September 2026 as part of the Qwen-Audio 3.1 suite, which cut realtime API prices by approximately 85 percent from the previous generation. The model processes audio input and output in a single neural pass rather than chaining separate STT, LLM, and TTS components, lowering end-to-end latency and eliminating pipeline error accumulation. The realtime model covers 31 languages for live voice interaction, including 11 major international languages and 20 Chinese regional varieties. It includes emotional tone detection: when the model perceives a low or distressed speaker mood it slows pacing and softens tone in response. Function calling is supported natively within the session, enabling tool invocations without breaking the audio stream. ASR first-character latency benchmarks at approximately 160 milliseconds. At approximately $0.018 per minute after the 2026 price cuts, Qwen-Audio-3.1-Realtime is among the most cost-efficient multilingual real-time voice APIs available through a major cloud provider. It suits Multilingual Product Builders shipping voice products to Chinese, East Asian, and European markets simultaneously, with the full Alibaba Cloud ecosystem available for downstream processing, storage, and telephony integration.