Google DeepMind

Genie 4

Research preview

Real-time world model that generates a scene with native audio, including ambient sound and dialogue, and accepts text prompts while the scene is running. DeepMind says Gemini 3.8 Flash watches the frames and breaks a goal into steps. Object permanence, legible text, and audio accuracy can still drift.

Latest update Oct 28, 2026

01

Current status

Trusted testers only as of Oct. 27, 2026. DeepMind says a later release is planned for Google AI Ultra in Project Genie and for early-access Gemini API and AI Studio partners. No price or end date for the preview. TechCrunch, citing a briefing, said scenes can stay consistent for up to two minutes; the DeepMind post says short sessions and does not state that duration.

  1. DeepMind releases Genie 4 as a research preview

    Google DeepMind on Oct. 27, 2026 released Genie 4, a world model with native audio and text steering during a running scene, as a research preview for trusted testers. A later release is planned for Google AI Ultra subscribers in Project Genie and for early-access Gemini API and AI Studio partners. DeepMind said Gemini 3.8 Flash breaks goals into steps and that object permanence, legible text, and audio accuracy can still drift; TechCrunch reported a two-minute limit from a briefing that the DeepMind post does not state.

    deepmind.google techcrunch.com + Alphabet

03

More from Google DeepMind8

04

More in AI Labs