Google DeepMind unveils Genie 4 world model
Google DeepMind has announced Genie 4, a new generation of its world-simulation model, though the material available so far names the team behind it without detailing capabilities, benchmarks or access.
Google DeepMind announced Genie 4 in a blog post titled "Genie 4: A new generation of world simulation." The release continues the lab's Genie line of world models.
The only part of the post available to us is the acknowledgements section. It does not describe what Genie 4 can do, how it was evaluated, or how it compares with earlier versions. It also gives no details on pricing, availability or which products, if any, will use it.
The acknowledgements show a large effort. They list 87 core contributors, including Jack Parker-Holder, Shlomi Fruchter, Tejas Kulkarni, Philip Ball, Inbar Mosseri and Aleksander Hołyński, plus a long list of cross-functional partners across Google. The post thanks senior leaders for guidance, among them Raia Hadsell, Aäron van den Oord, Josh Woodward, Koray Kavukcuoglu and Demis Hassabis. That is a sign of how much weight the company places on the project.
The post also thanks "trusted tester and early access users" who gave feedback. This suggests some outside users have had access to the model, but the material does not say who they are or whether wider access is planned.
We have not yet seen the technical claims, demos or evaluation results needed to judge how much Genie 4 improves on earlier versions. We will update this story when those details are available.