Model Card: Google DeepMind Genie 4
Genie 4 launched 2026-10-27 as a real-time interactive world model (not a text LLM). It is the first in the Genie line to natively generate ambient sound, music, and interactive dialogue, and the biggest leap is how characters move and react. Gemini 3.8 Flash runs in the generation loop as a director, breaking goals into steps and checking progress frame by frame. Access is trusted-tester-only for now, with Project Genie and Gemini API early access still coming soon; there is no public pricing, no API model ID, and no official benchmark scores. Object permanence and legible text remain known weak points.