Omnilingual ASR is a speech recognition system that supports over 1,600 languages, including many that lack previous ASR technology. It allows users to add new languages with minimal examples and no special skills. The system is designed for accessibility and includes various model options for different use cases.

This report presents the Qwen3-ASR family, featuring two advanced speech recognition models that support 52 languages. The 1.7B model offers top performance among open-source options, while the 0.6B model balances accuracy and efficiency, achieving rapid transcription and efficient forced alignment for text-speech pairs. Both models are released under the Apache 2.0 license for community use.