Qwen reports comparable or better WER than Whisper-large-v3 on several multilingual benchmarks (e.g. CommonVoice 10.77 → 9.18, MLS 8.62 → 8.55), though with some regressions on specific splits.

Link for the comparison: https://github.com/QwenLM/Qwen3-ASR?tab=readme-ov-file#evalu...