Google DeepMind unveils Genie 4, the world model that once tanked gaming stocks

The rollout of Google DeepMind’s last world model preceded a one-day drop of more than 20% in Unity’s stock. Genie 4 generates interactive environments in real time, this time with native sound and dialogue, more realistic physics and the ability to steer scenes with new prompts.

DeepMind announced Genie 4 on October 27, and it’s pitching the model as a research preview for trusted testers rather than a public launch. According to DeepMind’s own blog post, Genie 4 is natively multimodal: music, ambient sound and interactive dialogue all generate alongside the visuals, in the same pass, rather than getting bolted on afterward. Characters move and react on their own. You can also type a new instruction mid-scene, something like changing the weather or dropping in a new character, and the world adjusts without restarting.

That’s a real jump from Genie 3, which launched in August 2025 and could already generate navigable, text-prompted 3D worlds at 720p and 24 frames per second. Genie 3 held a scene together for a few minutes and modeled basic physics like gravity. It had no sound. Google acknowledged that Project Genie’s generated worlds did not always follow real-world physics. In May 2026, Google connected Project Genie to Street View, allowing users to generate interactive environments grounded in imagery from real locations. Genie 4 keeps the real-time interactivity while adding more realistic simulations of aerodynamics and collisions, plus what DeepMind describes as its most dramatic improvement: characters that move naturally, respond to their environments and can be steered in real time.

Here’s the part that makes this more than a research update. When Google widened access to Project Genie, the consumer interface built on Genie 3, back in late January and early February 2026, gaming stocks did not shrug it off. Clips circulated online of Genie-generated worlds that looked like GTA, Zelda and Doom, and investors reacted as if a level designer’s job had been automated overnight. Unity Software, the game engine company, fell as much as 22.5% to 24% in a single session, based on reporting from Sherwood News and Stockopedia. Roblox dropped between 12.8% and 13%. Take-Two Interactive, the parent of Rockstar Games, lost between 8% and 10.5%. Nintendo slid 4.79%, and CD Projekt fell 8.9%.

Frankly, that reaction looked overdone almost immediately. Unity CEO Matthew Bromberg posted on X and LinkedIn that Project Genie’s output was, in his words, unsuitable on its own for games requiring consistent, repeatable player experiences, because the frames are probabilistic: nobody, not the player, not the developer, knows exactly what happens next until it happens. Analyst Piotr Poniatowski at mBank called the sell-off unjustified. By the following Monday, Unity, Roblox and Take-Two had all clawed back at least 3% of the loss. Bloomberg’s coverage of today’s Genie 4 launch frames it plainly as an upgrade to the same model that rattled those stocks nine months ago, which means the market now has a precedent to measure this release against instead of reacting blind.

Genie 4 is not going out to the public yet. It’s rolling out as a research preview to trusted testers first, with DeepMind saying access is coming to Google AI Ultra subscribers through Project Genie, and to early partners through the Gemini API and Google AI Studio. That staged rollout matters. Project Genie’s earlier rollout capped generations at 60 seconds and offered only limited character control. It initially required the $249.99 Google AI Ultra subscription, but Google later introduced a $99.99 monthly Ultra tier and reduced its higher tier to $199.99. If Genie 4 ships to the public with those same limits intact just with better physics and sound layered on top, the gap between “demo that looks like a game” and “something you’d actually build a game in” stays wide.

Where this technology is already being tested outside entertainment is autonomous driving. Waymo has adapted Genie 3 into a world model that generates rare and unusual driving scenarios, including tornadoes, floods, wildfires and animals in the road, to evaluate the Waymo Driver before it encounters similar situations in the real world. The system can generate simulated camera and lidar data, giving Waymo another way to test long-tail driving challenges.

DeepMind hasn’t said when Genie 4 reaches the general public, or what developer access will cost once it does. Given how the market responded to the last version before most people had even tried it, those details will be closely watched by gaming investors.

Also read: Nvidia is in talks to lead a $20 million round for Simplismart • JPMorgan says lenders now demand proof AI firms can repay debt • A Crawler Labelled Roblox Downloaded 145 GB of 3D Models From Three Community Sites, Their Operator Says

This article is posted in AI News, check it out for more related stories.