All models
World modelsResearch access
Google DeepMind
Genie 3
Google DeepMind's research world model for generating photorealistic, controllable environments that can be explored in real time from a text description.
Plain-English overview
What Genie 3 actually is
Genie 3 generates a visual world as the user moves through it rather than rendering a fixed video in advance. DeepMind reports 720p output, 20–24 frames per second, and consistency over a few minutes of interaction.
It is best understood as an important research and experimental product, not a drop-in world API. Access is through Project Genie and does not come with public API pricing or a general integration contract.
Good fit for
- Understanding the frontier of interactive world generation
- Research demos and agent-environment exploration
- Evaluating Project Genie when eligible
Category comparison
The facts that matter for world models
These are provider-published specifications, not Cody benchmark scores. Follow the linked sources for current limits and endpoint-specific exceptions.
- Price basisPublic API cost where available; research and private-beta models may not publish one.
- Not published
- Access stagePublic API, research experiment, partner onboarding, or private beta.
- Research experiment / Project Genie
- What it createsThe kind of interactive environment or embodied experience produced.
- Realtime photorealistic navigable world
- InteractionPublished frame rate, latency, or realtime behavior with its original basis.
- 20–24 fps
- Visual outputProvider-published resolution or output format where available.
- 720p
- API accessWhether developers can integrate the model today and under what access model.
- No general developer API
API and provider access
Where to get Genie 3
Availability
Regions and access stage
Research/experimental access through Project Genie; no general developer API is documented.
Project eligibility and country availability can change and are not enumerated on the model page.
Check live availabilityData and training
The route matters.
No general developer API data policy is published for this research-access entry. Review Project Genie terms before uploading confidential prompts or reference material.
This is a concise reading of the cited provider material, not legal advice. A third-party gateway can have different storage, routing, training, and residency terms from the model maker's direct API.
Read the provider policyFrequently asked questions
Genie 3 FAQ
What is Genie 3?
Google DeepMind's research world model for generating photorealistic, controllable environments that can be explored in real time from a text description. Genie 3 generates a visual world as the user moves through it rather than rendering a fixed video in advance. DeepMind reports 720p output, 20–24 frames per second, and consistency over a few minutes of interaction.
When was Genie 3 released?
Genie 3 was released on August 5, 2025 according to the cited provider materials.
Where can I access Genie 3?
Research/experimental access through Project Genie; no general developer API is documented. The access routes listed in this guide are Google DeepMind.
How much does Genie 3 cost?
Not published. Genie 3 is presented through Project Genie rather than a generally available metered developer API.
Where is Genie 3 available?
Research/experimental access through Project Genie; no general developer API is documented. Project eligibility and country availability can change and are not enumerated on the model page.
Is my Genie 3 API data used for training?
No general developer API data policy is published for this research-access entry. Review Project Genie terms before uploading confidential prompts or reference material. The policy belongs to the provider route and account terms, so verify it again before production use.
Research trail
Sources for this guide
Keep comparing
More world models models
World Labs · Stable
Marble 1.1 Plus
World Labs' public API model for turning text, images, panoramas, multi-view sets, or video into large explorable 3D worlds and spatial assets.
Read guideDecart · Preview
Oasis 3
Decart's action-conditioned world model for physical-AI training, generating synchronized, controllable simulation views in real time.
Read guideOdyssey · Private beta
Odyssey-2 Max
Odyssey's largest general world model, focused on real-time next-state simulation, physical accuracy, open-ended futures, and longer interactive rollouts.
Read guide