DeepMind's Genie 4 adds native audio and Gemini-powered reasoning

Google DeepMind released Genie 4 as a research preview for trusted testers, the first real-time world model to generate audio in the same forward pass as visuals. Gemini 3.8 Flash acts as a continuous guiding agent, breaking user goals into steps and steering the simulation. No timeline for wider rollout was announced.

• Genie 4 generates audio and video in a single forward pass, not a separate pipeline
• Gemini 3.8 Flash continuously steers the simulation and verifies each step
• Available only to trusted testers, with no firm rollout timeline
• Demo showed accurate physics: watermelon splitting and maple seed autorotation