World models
that see and hear.

We compress audio-visual reality into models that predict and generate it.

An independent lab working from first principles on long-horizon, controllable simulation — efficient enough to build on hardware you can actually buy. Learning and compression are the same problem; we build at that seam.

Focus