WanSong

Tongyi/Wan's ~25B single-stage diffusion long-form music generator: five-minute multilingual songs with dual-stem (vocals + BGM) output; claims 7.43% pronunciation error vs Suno V5's 22.8%. No open weights yet — read alongside Wan-Dancer and Wan-Streamer as the build-up to a "Wan 3.0" open video/audio wave.