Model
1 entry · 17 March 2026
The company said it rewrote its codebase from Google TPU infrastructure to a GPU-native PyTorch stack, enabling native 2K output at roughly five times the generation speed of V7.
Models & capabilities