TLDRocket
27 October 2026
Google DeepMind’s Genie 4 grabbed the center of the AI day by making the “world model” feel less like a slideshow and more like something you can steer. In a research preview, Genie 4 adds audio and tighter control loops, letting prompts redirect what happens mid-simulation. DeepMind says it can keep a consistent scene for up to two minutes—double Genie 3’s limit—while supporting real-time steering and more coherent physics, motion, and audio. The rollout is staged: trusted testers first, then availability for Google AI Ultra subscribers through Project Genie, with earlier access for partners and the Gemini API and AI Studio.
Read the full briefing →