Overview
- Genie 4 was announced on Tuesday, October 27, 2026, as a research preview that DeepMind has opened to trusted testers with a phased rollout planned for Google AI Ultra, Project Genie, Gemini API users and Google AI Studio partners.
- The system generates interactive, high‑fidelity virtual worlds that produce native, time‑aligned audio and accept real‑time text prompts to change weather, camera angles, characters or objects during a running simulation.
- DeepMind says Genie 4 can keep coherent scenes for up to two minutes, shows improved physics and more natural character behavior, and is designed to work with Gemini models that plan tasks and direct simulations.
- Engineers warned the model still has practical limits that block production use today, including detail drift over long runs, imperfect audio accuracy, and poor legibility of text that appears inside the simulated world.
- The launch deepens competition in the emerging world‑model market as rivals and recent deals increase consolidation, and it could speed new workflows in games, interactive learning, film prototyping and robotics testing.