AI and security briefsModels

What is Google DeepMind's Genie 4, and can my team use it today?

By Brandon Da Costa, founder of Digiton Dynamics

Published 28 October 2026Updated 28 October 2026Editorial policy

Genie 4 is a world model that turns text prompts and keyboard input into an explorable, interactive video world with sound. Google DeepMind announced it on 27 October 2026 as a research preview for trusted testers. There is a waitlist and no public API, so a team can plan an evaluation but cannot build on it today.

Who should act: Game, product and simulation teams, and anyone who trains or tests agents or robots, who want to know whether Genie 4 can enter a pipeline this quarter.

What changed

On 27 October 2026 Google DeepMind announced Genie 4, which it calls its new world model, with better realism and real-time interactive simulation (Google DeepMind blog, 27 October). It is natively multimodal. Music, ambient noise and interactive dialogue are generated together with the world. DeepMind calls it its first real-time world model to generate audio natively.

You steer a running world with standard keyboard controls for navigation and camera, and with text prompts that change the weather, move the camera or add a person or an object (Google DeepMind blog, 27 October). DeepMind says Genie 4 builds on the Gemini family. Gemini Omni anchors the simulation in real-world appearance and behavior, and other Gemini models help keep it coherent.

DeepMind also describes a guiding-agent setup, in which Gemini 3.8 Flash watches the frames Genie 4 generates, splits a goal into steps and checks each one. A second demonstration has Apptronik's Apollo humanoid playing checkers (Google DeepMind blog, 27 October). The post gives no success rate for either.

Access is narrow. DeepMind says Genie 4 is a research preview for trusted testers and is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio. It points to a waitlist for Project Genie. Every generated video and audio track carries a SynthID watermark.

Who it affects

• Game and prototyping teams. DeepMind says developers tested it to go from an idea to an explorable prototype in seconds (Google DeepMind blog, 27 October).
• Film and journalism teams. Testers used it for pre-visualization and to turn one archival photograph into a simulated space.
• Robotics and agent teams. DeepMind lists training environments for robots and agents among its goals. The post shows no agent trained or tested in Genie 4 with a result.
• Everyone who builds on the Gemini API. Nothing changes yet. Google's Gemini API model list and changelog, read on 28 October, name no Genie model.

What to do now

1. Join the waitlist for Genie 4 in Project Genie from the DeepMind page. If your company has a Google contact, ask whether you can be an early access partner for the Gemini API and Google AI Studio.
2. Ask for what the post does not give: a model ID, the price, rate limits and maximum session length.
3. Write your acceptance tests before access arrives. DeepMind says detail and object permanence can drift in longer sessions, after many prompt edits and while moving around the scene. Test exactly that.

   scene 01: place a red cube, explore for 60 s, is it still where it was?
   scene 02: show a sign with a 12-character word, is it legible?

4. Keep it out of CI and training loops for now. The post describes no API and no way to replay a world, so a test built on it cannot be repeated.
5. Check the Gemini API model list now and again after Google's next announcement.

   curl -s "https://generativelanguage.googleapis.com/v1beta/models?key=$GEMINI_API_KEY" | grep -i genie

What to watch

• A Gemini API model ID, a price and rate limits. None are published as of 28 October, and "coming soon" has no date.
• Independent tests. The post is DeepMind's own demonstrations and tester feedback. It reports no benchmark and no measured session length.
• The limits DeepMind lists itself: complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency are described as active research.

Questions people ask

Digiton's production review covers how a new model gets approved before it touches production systems, and the checks are on the vibe coded app audit page.

How this brief was made: researched from the sources below, drafted with AI assistance and checked against each source on the dates shown. See the editorial policy.

Sources

1. Google DeepMind, Genie 4: a new generation of world simulation, 27 October 2026, read 28 October 2026. https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
2. Google DeepMind, Genie model page, read 28 October 2026. https://deepmind.google/models/genie/
3. Google AI for Developers, Gemini API models, read 28 October 2026. https://ai.google.dev/gemini-api/docs/models
4. Google AI for Developers, Gemini API changelog, read 28 October 2026. https://ai.google.dev/gemini-api/docs/changelog

More briefs

How these briefs are sourced and corrected. All AI and security briefs