Gen0

GEN-0 marks the arrival of scaling laws in the physical world.
For years, language models proved that cognition scales predictably with data and compute. Now, embodied foundation models are extending that principle into matter itself.

Generalist AI’s GEN-0 is trained on over 270,000 hours of real-world manipulation data — not text, but touch, friction, and motion. Its new “Harmonic Reasoning” architecture lets robots think while acting, blending perception and control into one continuous loop. The outcome is fluid, human-like motion and an ability to generalize across robots with 6, 7, or even 16+ degrees of freedom.

The research reveals a profound threshold: true physical intelligence emerges only beyond 7B parameters. Smaller models ossify — unable to absorb the complexity of embodied experience.

This is computation meeting physics — the moment where kinetic semantics becomes real.
GEN-0 doesn’t just automate motion; it formalizes the laws of physical learning. The next frontier in AI is not language, but continuity — the intelligence of matter itself.