DEPLOYDatabase

Brain

Genie 3

Google DeepMind's world model that learns interactive action spaces from video without labeled data. Generates playable environments for robot training. Closed API only as of 2026. Builds on the Genie family of latent action models.

World model · Maturity: Research · Closed

Machine-readable surfaces

Architecture

Latent action model, learns compressed action space from video

Key facts

Genie 3 announcement (primary DeepMind)
On Aug 5 2025 Google DeepMind announced Genie 3, a general-purpose world model. The post says a text prompt can generate dynamic worlds navigable in real time at 24 frames per second, retaining consistency for a few minutes at 720p, with visual memory extending as far back as one minute. It was announced as a limited research preview for a small cohort of academics and creators, not a public product (Google DeepMind)
Project Genie access (primary Google)
On Jan 29 2026 Google said it was rolling out Project Genie, an experimental research prototype powered by Genie 3 (plus Nano Banana Pro and Gemini), to Google AI Ultra subscribers in the U.S. who are 18 or older, not including Google AI Ultra for Business. The post lists known limits including worlds that may not match real-world physics and generations limited to 60 seconds. Frame access and limits as Google-stated for the prototype, not the August research preview (Google)
General-purpose world model
Google DeepMind describes Genie 3 as a general-purpose world model that generates photorealistic, interactive environments from text descriptions and lets users explore them in real time (Genie 3).
Interactive generation limits
Google DeepMind's launch post says Genie 3 can generate dynamic worlds at 24 frames per second and 720p while retaining consistency for a few minutes, and was initially offered as a limited research preview (Genie 3 launch).
Genie interactive worlds
Google DeepMind says Genie 3 generates interactive environments from text prompts that users can navigate in real time (Genie 3).
Genie real-time output
Google DeepMind reports that Genie 3 can generate navigable worlds at 24 frames per second and 720p resolution for several minutes (Genie 3).
Genie embodied research
Google DeepMind says Genie 3 worlds were used to test embodied-agent research and can support agent navigation goals (Genie 3).
Real-time interactive world model announcement (DeepMind)
On August 5, 2025 Google DeepMind announced Genie 3 as a general-purpose world model that can generate interactive environments from text prompts, stating real-time navigation at 24 frames per second and 720p with consistency for a few minutes. Treat performance claims as company-stated (Google DeepMind).
Limited research preview and agent training framing (DeepMind)
DeepMind stated Genie 3 was released as a limited research preview for a small cohort of academics and creators, and described using Genie 3 worlds with a recent SIMA agent for embodied goal-directed navigation research. Treat availability and agent results as company-stated (Google DeepMind).

Developed by (1)

Common questions

What is Genie 3?
Google DeepMind's world model that learns interactive action spaces from video without labeled data. Generates playable environments for robot training. Closed API only as of 2026. Builds on the Genie family of latent action models.
Who developed Genie 3?
Genie 3 is credited to Google DeepMind on the DEPLOY registry. Each developer attribution is verified via primary sources.
Is Genie 3 open source?
No. Genie 3 is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.
What type of AI is Genie 3?
Genie 3 is a world model, built on a Latent action model, learns compressed action space from video architecture on the DEPLOY registry.
What is Genie 3's maturity stage?
Genie 3 is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.
Which robots run on Genie 3?
No robot models on the DEPLOY registry are recorded as running Genie 3. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.

Sources (3)

  1. https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/
  2. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie/
  3. https://deepmind.google/models/genie/
Methodology: Verified · 3 sources (no primary) · last reviewed 2026-10-09

Verification posture

Verified

Low confidence

Review state

Stable

Last reviewed 2026-10-09

Maturity + lifecycle

Maturity stage: research

Sources by quality tier

3
unclassified
Unclassified source

The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.

Methodology surface for Genie 3.

Canonical ID 97239e0f-b2b4-4173-9758-a50f48732845