What happened

On 27 October 2026 Google DeepMind announced Genie 4, which it describes as a real-time world model: you move through a generated scene with standard keyboard controls and steer it with text prompts while it runs. DeepMind calls it “natively multimodal” and says it is its first real-time world model to generate audio natively, meaning ambient sound and dialogue are produced together with the visuals.

Genie 4 is a research preview for trusted testers. DeepMind says it is “coming soon” to Google AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, and it links a waitlist.

Key details

  • Gemini in the loop: according to DeepMind, Gemini Omni grounds the visuals in how the real world looks and behaves, and other Gemini models help keep scenes coherent. A Gemini model, which the post names as Gemini 3.8 Flash, acts as a guiding agent during generation: it watches frames, breaks a goal into steps and adjusts its plan when the user intervenes.
  • What DeepMind claims: improved photorealism, more fluid movement and more realistic physics, such as aerodynamics and collisions. These are DeepMind’s own descriptions; the post publishes no benchmarks.
  • Not published: the blog post gives no resolution, frame rate, session length, pricing or list of supported countries. For comparison, Unite.AI reports that Genie 3 ran at 24 frames per second and 720p with consistency lasting a few minutes.
  • Safety: DeepMind says Genie 4 has safety guardrails, went through internal safety evaluations before the preview, and that all generated video and audio carries SynthID watermarking.
  • Stated limitations: visual details and object permanence “can still shift or drift over time”, especially in long sessions or after several edits. DeepMind also lists complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency as open research areas.
  • Who is testing: Unite.AI reports that testers in creative industries, journalism, education, gaming and engineering are already evaluating it. Bloomberg’s headline links the release to the earlier sell-off in video game stocks that followed Project Genie; we have not independently checked any market movement.

Why it matters

World models generate an interactive environment rather than a fixed video clip, which is why they interest game developers, simulation and robotics teams, and anyone who prototypes scenes. Adding native audio and a Gemini agent that plans toward a goal moves Genie from a visual demo toward something closer to a playable prototype.

For most readers nothing is available yet: access is limited to trusted testers, the Ultra and API tiers are only “coming soon”, and DeepMind offers no measurable quality figures to compare against Genie 3 or other world models. Treat the announcement as a capability preview, and wait for independent hands-on tests of drift, latency and audio quality before drawing conclusions.

Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article. Google DeepMind is a competitor of Anthropic. All capability claims are DeepMind’s unless stated otherwise.

Sources

  1. Genie 4: A new generation of world simulation Primary source , Google DeepMind
  2. Google DeepMind Rolls Out Genie 4 World Model in Research Preview Unite.AI
  3. Genie 4 Expands Google's AI Capabilities in Interactive Virtual Environments Bloomberg