← All models

Genie 3

Published Publisher Google

Positioning and release date

Genie 3 is a general-purpose world model developed by Google DeepMind for generating interactive, dynamic environments. The Genie 3 research preview was officially announced on August 5, 2025, initially offering limited early access to a small number of academic researchers and creators. (deepmind.google)

Capabilities

  • Generates dynamic worlds from text prompts that can be navigated in real time; the officially demonstrated generation specifications are 720p at 24 frames per second.
  • Environments can remain broadly consistent for several minutes, with visual memory extending to approximately one minute earlier.
  • Can simulate natural phenomena such as water and lighting, as well as interactions with the environment; in addition to navigating, users can trigger world events through text, such as changing the weather or introducing objects and characters.
  • The SIMA agent was used by Google DeepMind to test the ability to perform navigation and goal-directed tasks in generated environments. (deepmind.google)

Uses

It can be used to explore interactive generative media and to research agent training and evaluation; Google DeepMind also proposed potential uses including education, skills training, and simulation for robotics and autonomous systems. These uses do not mean that its suitability for real-world deployment has been verified. (deepmind.google)

Limitations

  • The range of actions agents can perform directly is limited; simulating complex interactions among multiple independent agents in a shared environment remains a research challenge.
  • Geographic accuracy for real places cannot be guaranteed; clear text generally needs to be provided in the input world description.
  • The original announcement stated that continuous interaction could last for minutes, not hours; generated environments may also deviate from the laws of real-world physics.
  • Model capabilities are not equivalent to product features: the Project Genie prototype introduced by Google in January 2026 limited each generation to 60 seconds and did not at the time include the ability to change the world through prompts during exploration. (deepmind.google)

Access and pricing

In January 2026, Google launched Project Genie, an experimental prototype based on Genie 3, Nano Banana Pro, and Gemini, initially available to Google AI Ultra subscribers aged 18 or older in the United States. A Google help page states that generating worlds in the early-access prototype does not consume AI credits; this is a product access condition and is not equivalent to standalone API pricing for the Genie 3 model. (blog.google)

As of November 18, 2026, the official price, billing currency, input charges, and output charges for a standalone Genie 3 API have not been verified, nor has an official API model ID; this does not imply that it is free.

Official sources