Google DeepMind has unveiled Genie 4, a new world model that the company says improves on Genie 3 in almost every way. The biggest changes are more realistic physics and more realistic humans inside the virtual worlds it creates. The reveal follows Genie 3, which last year sent gaming stocks tumbling when it was first shown.
It is easy to mistake the output of Genie 4 for a video game. That is exactly why Genie 3 caused so much worry in the gaming industry. But DeepMind does not see it that way.
What Is a World Model and Why DeepMind Built Genie 4
According to Google, world models are a key element in the march toward artificial general intelligence (AGI). A world model creates a virtual world, but it does not require pre-rendering environments. In simple terms, the model builds the world as it goes instead of playing back a finished scene.
That difference matters. Pre-rendered environments are fixed — someone has to build them in advance. A world model generates the environment on the fly, which is why the results can look like a game even though the system was not built as one.
DeepMind researchers have been working on this updated model since Genie 3. The company is not resting on its earlier success, and Genie 4 is the result of that work.
Genie 4 Access: Paid Use May Come Soon
There is a practical side to the announcement. If you pay Google enough money, you might actually get to use Genie 4 before too long. That suggests DeepMind is thinking about commercial access, not just research demos.
The original Genie 3 reveal was enough to move gaming stocks. That reaction shows how seriously the industry takes world models — a system that can generate interactive worlds is seen as a direct challenge to how games are made.
- Genie 4 improves on Genie 3 in almost every way, according to the company
- The new model promises more realistic physics and humans
- DeepMind frames world models as a step toward AGI, not as game engines
- Paid access to Genie 4 may arrive before long
Our Take: Genie 4 Is a Research Milestone With a Business Plan
In our view, the most telling part of this story is not the graphics. It is the framing. DeepMind is careful to say Genie 4 is a world model aimed at AGI, not a game product. That distinction protects the research story, but the mention of paid access tells you Google also sees a market here.
To put it plainly, the gaming industry's fear after Genie 3 was not irrational. A system that builds virtual worlds without pre-rendering environments changes the cost of making those worlds. If Genie 4 delivers on realistic physics and humans, the gap between a generated world and a designed one gets smaller.
Readers should watch two things. First, whether DeepMind shows real, verifiable output from Genie 4 rather than short clips. Second, what the paid access actually looks like — who gets it, and at what price. Until then, treat the "almost every way" claim as a promise, not a proven result.
The bigger point is the direction of travel. World models are now a major focus of AI labs around the world, and Google is pushing to stay ahead. Genie 4 is less about one product and more about a race — and that race is now moving toward more realistic physics, more realistic humans, and eventually, paying customers.