HomeFull speed ahead. Keep track of advancements in every frontier

TheTechno Optimist.

Genie world modelsGoogle DeepMind—VeoGoogle DeepMind—GeminiGoogle DeepMind—Tensor Processing Unit (TPU)Google DeepMind—AlphaFoldGoogle DeepMind—Genie world modelsGoogle DeepMind—VeoGoogle DeepMind—GeminiGoogle DeepMind—Tensor Processing Unit (TPU)Google DeepMind—GeminiGoogle DeepMind—
Genie world models
Courtesy of Google DeepMind · Press kit, editorial use (opens deepmind.google)

Genie world models

by Google DeepMindGBUS

PilotStage 3 of 5

Genie 3 is in limited public use through Project Genie, for Google AI Ultra subscribers.

Updated 19 May 2026Checked 25 Sep0 updates this week

Milestones

No announced next step
  1. Genie 1 makes playable 2D worlds from game videosFeb 2024Complete.
  2. Genie 2 generates 3D worlds from a single image4 Dec 2024Complete.
  3. Genie 3: real-time worlds at 720p and 24 frames per second5 Aug 2025Complete.
  4. Project Genie opens to Google AI Ultra subscribers29 Jan 2026Complete.

Most important updates

  • 19 May 2026
  • 29 Jan 2026
  • 5 Aug 2025
  • 4 Dec 2024
  • 23 Feb 2024

Current obstacles

  • Keeping worlds consistentEach frame is predicted from earlier ones, so small errors pile up. Worlds hold together for minutes, not hours.
  • Compute costDrawing a world live for each user takes a lot of computing power, so sessions are kept short.

Physics limits

  • Mistakes snowball frame by frameEach frame is predicted from earlier ones, so small errors feed back and grow. Genie 3 stays consistent for minutes, not hours.
  • Real time leaves milliseconds per frameAt 24 frames per second the model has about 42 ms to make each 720p frame (0.9 million pixels) and react to input, which caps how large the model can be.
  • Physics is imitated, not computedThe model copies what video usually looks like; no physics equations are solved, so objects can pass through each other or appear from nowhere.

How it works

4 parts
Google's data centre in The Dalles, Oregon: Genie is trained on Google's TPU fleet in buildings like this
Google's data centre in The Dalles, Oregon: Genie is trained on Google's TPU fleet in buildings like thisPhoto: Tony Webster · CC BY 2.0 (opens commons.wikimedia.org)
Frames

One frame at a time

Genie predicts each new video frame from the frames before it and the user's latest key press, so the world is generated as you move.

Actions

Learned controls

Genie 1 learned what 'actions' are from unlabelled gameplay videos, inferring the move that best explains the change between frames.

Memory

Remembering the world

Genie 3 keeps scenes consistent for a few minutes: look away and back, and objects are still there, at 720p and 24 frames per second.

Uses

Training grounds for agents

DeepMind uses generated worlds to train and test AI agents and robots on situations too rare or risky to set up for real.

Update log

5 updates

Tue 19 May

  • Minor: BlogSoftware

Thu 29 Jan

  • Major: BlogSoftware

Tue 5 Aug 2025

  • Major: BlogResearch

Wed 4 Dec 2024

  • Major: BlogResearch

Fri 23 Feb 2024

  • Minor: PaperResearch

About Google DeepMind

The team behind Genie world models

Google DeepMind

Gemini, AlphaFold, Genie, Veo, TPU chips

Google DeepMind is Alphabet's AI lab, formed in 2023 from DeepMind and Google Brain. It builds the Gemini models and leads in AI for science. AlphaFold won a share of the 2024 Nobel Prize in Chemistry.

  • Gemini 3 (Nov 2025) led many benchmarks. Gemini 3.5 Pro, announced in May 2026, is delayed.
  • AlphaFold predicted structures for over 200 million proteins and won a share of the 2024 Chemistry Nobel.
  • Google's own Ironwood TPU chip is used in-house and sold to outside labs, including Anthropic.

deepmind.google (opens deepmind.google)Company profile →

Founded
201016 yrs
Headquarters
United KingdomUnited States
Status
Subsidiaryof Alphabet
Valuation
Alphabet-owned
Works in
AIFrontier modelsComputingBiotech
Coverage
5 programs · 30 updateslatest 29 Jul 2026checked 25 Sep
Customers
Anthropic
Partners
Isomorphic LabsBroadcom
Acquired by
Alphabet
People
Demis HassabisCEO and co-founderShane LeggCo-founder and Chief AGI ScientistJohn JumperAlphaFold lead; 2024 Nobel laureate
Primer