General Intuition Raises $220M at $6.2B for World Models

General Intuition Inc. has closed a $220 million funding round that values the company at $6.2 billion. The startup builds world models, AI systems that generate synthetic video which can be used to train other AI models, including those that control robots.

The round brings in a long list of investors: Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst. It lands roughly a year after General Intuition was spun out of Medal B.V., a video sharing startup.

A gaming clip library as training data

The company's origin explains its data advantage. Medal offers a free tool that lets players capture and record footage from video games. General Intuition uses that footage to train its own AI models.

That matters because training data is the bottleneck for many AI projects. Developers teach a model by showing it examples of the work it will eventually do. A system meant to drive a robotic arm, for instance, might learn from video of robotic arms welding car parts. Recording that kind of footage by hand is slow and expensive.

General Intuition's pitch is to skip much of that manual work. Its world models produce synthetic footage that developers can feed into their own training pipelines.

What MIRA does differently

The company's latest model, MIRA, came out in June. General Intuition says it improves on earlier models in a few specific ways.

1. Length. Conventional video generators can only produce clips lasting a few seconds or minutes. That rules them out for simulating long factory automation workflows, which is exactly what robot developers often need. General Intuition says MIRA can "run infinitely without diverging."

2. Motion. MIRA can render complex scenes in which several fast-moving objects interact. Simulating movement this way is useful for jobs such as teaching robots to avoid collisions.

3. Efficiency. According to the company, MIRA renders 20 frames per second at 720 by 576 pixels on a single B200 graphics card.

Two design choices drive that efficiency. The first is latent diffusion. Instead of working directly on video frames, MIRA performs its calculations in a latent space, a compressed representation of the frames that takes up less memory. A smaller memory footprint means less processing time. The second is size: MIRA has just 5.6 billion parameters, only a fraction of what frontier models contain.

Not a product yet

There is an important caveat. MIRA is a research demonstration, not a commercial data generator. In its current form it can only produce footage of a single video game. General Intuition still describes it as a "stepping stone to physical AI," pointing to its efficiency and rendering quality.

A commercial version is already in testing with a small group of customers working on robotics, simulation and entertainment. Alongside the funding news, the company opened a waitlist for that offering. It plans to spend the new capital on hiring more AI researchers.

Our Take

A $6.2 billion valuation for a company whose flagship model currently renders one video game says a lot about where investor attention is heading. World models have moved from a research niche to a funding category of their own. General Intuition's round follows closely on AMD's purchase of World Labs, another bet that simulated environments will become core infrastructure for robotics. Taken together, this suggests the industry increasingly sees synthetic data as a way around the scarcity of real-world robot footage.

General Intuition's specific edge is unusual: a pipeline of gaming footage inherited from Medal. It is worth asking how well skills learned from game worlds transfer to factory floors and warehouses. The company's own framing, a "stepping stone," is an honest admission that this gap has not yet been closed.

The efficiency claims deserve attention too. A 5.6 billion parameter model running on a single GPU could make synthetic data generation cheaper and more accessible than approaches built on frontier-scale systems, if the results hold up outside a single game.

What to watch next: whether MIRA's successors can render varied, realistic environments, which customers move from the waitlist into paid deployments, and whether this round, like other recent large raises such as EliseAI's $350 million, turns into measurable commercial traction rather than just a higher price tag.