General Intuition says it raised $220M at a $6.2B valuation

Why it matters

General Intuition's new valuation puts a large price on a specific technical wager: that gameplay data with recorded player actions can help train agents for physical environments before those models are broadly available.

General Intuition said in a thread on X that it raised another $220 million at a $6.2 billion valuation, as it begins making its action and world models available to partners. The company named Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst as investors. The thread does not specify whether the valuation is pre-money or post-money.

https://x.com/gen_intuition/status/2104901412987748860

For Pim de Witte, General Intuition's co-founder and CEO, the financing extends a bet rooted in the gaming company he built before it. De Witte co-founded Medal, a platform for capturing and sharing gameplay clips, and General Intuition spun out of Medal in 2025. Before that, he built and ran a large private RuneScape server as a teenager and worked in humanitarian technology, including with Médecins Sans Frontières, according to TechCrunch's reporting and a profile interview.

The new round follows a $320 million financing announced on June 25th at a $2.3 billion valuation. That round came after General Intuition's $133.7 million seed in October 2025, led by Khosla Ventures and General Catalyst. Those earlier disclosed rounds totaled $453.7 million; counting the new $220 million as closed would bring disclosed funding to about $673.7 million. The valuation in the X announcement is nearly three times the June figure, though the basis for comparing the two valuations is not specified. In August, TechCrunch reported that General Intuition was finalizing a round at a $6 billion pre-money valuation; the new post says it has now closed at a $6.2 billion valuation.

General Intuition's pitch is that Medal supplies something many AI training datasets lack: video paired with records of the player's inputs. Those action labels capture which controls a person used and when, allowing a model to connect what it sees with what happened after an action. In its thread, General Intuition says Medal is on track to upload three billion videos a year and claims its training capacity exceeds what leading robotics and world-model efforts can use. That scale and the claim about competitors come from the company.

The company wants to use the data to train two related systems: action models that choose what to do, and world models that predict what follows. Its website describes the world model as part of the training process and says General Intuition has onboarded initial commercial partners in games, simulation and robotics. The website also says those partners will get selective access ahead of a broader release. The partner sign-up page currently invites prospective users to join an early-access list; it describes the models as coming soon.

A public demonstration offers a more concrete measure of what the approach can do so far. General Intuition and Kyutai, working with Epic Games, built MIRA, a playable model of Rocket League trained on about 10,000 hours of synthetic gameplay. The model generates a four-player game at 20 frames per second in response to player inputs. MIRA shows a model learning to simulate a specific game environment. It does not establish that a model can reliably control a robot in unfamiliar physical settings, the broader transfer claim behind General Intuition's fundraising pitch.

That gap is the central risk behind the valuation. In June, TechCrunch observed a General Intuition model navigating a game and a quadruped robot moving around the company's office after a small amount of real-world fine-tuning. The report also noted that dependable physical-world performance at scale remains unsolved across this class of systems. General Intuition is asking investors to fund the possibility that gameplay can make physical-action training cheaper and more varied; the new valuation prices that possibility well ahead of broad model availability.

The company's own thread says it will expand hiring in New York and Europe. It is also using the round to move from demonstrations toward access for external partners. Whether gameplay-derived action labels transfer across games, simulations and physical machines is the technical question those partners will test. For now, the company has named the backers and opened the waitlist, while its broad model release remains ahead.