Qwen3-ASR

Qwen3-ASR is Alibaba's Qwen team's open-source automatic speech recognition model family, released on January 30, 2026. It covers 30 languages plus 22 Chinese dialects, along with English spoken with accents from multiple countries and regions, which is the coverage gap most open ASR models leave open.

The family targets both realtime and offline transcription workloads, with features like hot words, speaker separation, punctuation restoration, and multilingual plus dialect recognition. The companion Qwen-Audio-3.1-ASR-Flash-Filetrans API variant handles offline long-audio transcription. Open weights make it deployable wherever transcription needs to run cheaply and privately.