Mochi 1 Review — Genmo's 10B Open-Source Video Model and the AsymmDiT Architecture That Proved Motion Quality Was Solvable

Mochi 1 (Genmo, October 2024) is a 10-billion-parameter open-source text-to-video model built on AsymmDiT — an asymmetric multimodal diffusion transformer that allocates 4× more parameters to visual than text processing. At release it was the largest open video model ever, and it set the community benchmark for smooth, physically coherent motion in fluid dynamics, hair, cloth, and human movement. Apache 2.0. 480p, 30 fps. Rating: 4/5.