General Intuition raises $220M at $6.2B valuation for world models

General Intuition builds world models that generate synthetic video for developers training robots and other AI systems. The Medal spinout has raised $220 million at a $6.2 billion valuation, according to SiliconANGLE, which reported that the company announced the financing on Sept. 29.

A world model is meant to simulate how a scene changes as objects move and interact. Instead of recording every robotic arm, collision or factory sequence in the physical world, a developer could generate examples inside the model and use them for training. General Intuition is testing the commercial technology with a limited group working in robotics, simulation and entertainment.

Its starting data came from Medal, whose free software lets players capture and record game footage. General Intuition uses those recordings to train its models, connecting the new company’s research to an existing flow of actions in interactive environments even as its intended uses reach beyond games.

The newest model, MIRA, shows the technical pitch. General Intuition says the 5.6-billion-parameter system can keep generating coherent scenes, handle multiple fast-moving objects and render 20 frames per second at 720 by 576 resolution on one Nvidia B200 graphics card. It uses latent diffusion, which performs calculations on compressed representations rather than full video frames.

But MIRA is still a research demonstration and currently generates footage from only one video game. The $6.2 billion valuation therefore arrives before General Intuition has turned that narrow demonstration into a broadly available product: it opened a waitlist alongside the financing and is testing a commercial version with a limited set of customers.

Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst provided the capital. General Intuition said it plans to use the money to hire more AI researchers.