Alibaba Qwen launches Qwen-Audio 3.1 family with up to 95% ASR price cut and TTS-Next diffusion model
Alibaba's Qwen team launched the Qwen-Audio 3.1 family of five models at the Apsara Conference on September 23, with ASR prices cut by up to 95%, TTS by about 70%, and Realtime by about 85%, news of which was widely circulated on September 25.
At the Apsara Conference on September 23, Alibaba's Qwen team launched the Qwen-Audio 3.1 family with five models in a single release. The base ASR supports 30 languages and 16 Chinese dialects with first-character response around 160 milliseconds, and natively cleans filler words and repetitions at the model layer. TTS adds cross-language timbre transfer, with emotion, speed, and style now controllable through natural-language prompts. The Realtime model is full-duplex, supporting simultaneous speaking and listening with arbitrary interrupts.
Two models are new. ASR-Next adds speaker labels, timestamps, emotion recognition, and identification of ambient sounds and machine noise alongside transcription. TTS-Next unifies a language model with a diffusion process to generate voice, sound effects, and background audio in a single inference pass, targeting audiobooks, podcasts, games, and ads. In parallel, Qwen3.8-LiveTranslate reduced simultaneous interpretation latency from 2.8 to 2.3 seconds, landing within the 2-to-6-second ear-voice-span typical of professional human interpreters.
Pricing cuts were aggressive across the lineup: ASR up to 95%, Realtime about 85%, TTS about 70%. At the Beijing node, Qwen-Audio-3.1-ASR-Flash is 0.8 yuan per million input tokens and 2.7 yuan per million output tokens. At 25 tokens per second, this works out to roughly $0.016 per audio hour, about one-ninth of Meta Muse Voice Transcribe at $0.18 and roughly one-nineteenth of Gemini 3.5 Transcribe at $0.30. The metering unit shifted from audio seconds to tokens on the same day, so the savings scale with speech density. The Realtime model is currently tied to Alibaba's own Agent products including Qoder and Tongyi Office, plus QwenNote, A2, Eva, and the Qwen AI glasses. Alibaba also open-sourced the Qwen-Audio-Agent framework for real-time voice agents.