world-models/world-model-registry

World Model Registry

World Model Registry is a curated, source-linked dataset for mapping the world-model ecosystem across Physical AI, robotics, learned simulation, planning and predictive representations.

Version: 0.1.0
Last verified: 2026-09-24
Initial records: 10

Why this registry exists

The label world model now covers several technical patterns:

World Foundation Model
World Action Model
Interactive World Model
Planning World Model
Action-Conditioned Video World Model
Predictive Representation / World-Model Backbone

The registry makes those differences explicit instead of placing every model in one undifferentiated list.

Core fields

• repo_id
• organization
• model_name
• world_model_type
• architecture_family
• modalities
• domain
• action_conditioned
• interactive
• prediction_target
• planning_support
• control_support
• robotics
• physical_ai
• spatial_reasoning
• simulation
• synthetic_data
• parameter_count_b
• weights_available
• access_status
• license
• commercial_use
• base_model
• paper_id
• verification_status
• last_verified
• source_urls
• notes

Classification principles

USE THE MODEL CARD
LINK THE SOURCE
SEPARATE DOCUMENTED FACTS FROM INTERPRETATION
MARK GAPS INSTEAD OF GUESSING

A model is only marked action-conditioned, interactive, planning-capable or control-capable when its documentation supports that property.

Initial scope

The first release includes examples of:

• NVIDIA Cosmos Predict-2 world foundation models
• Cosmos Policy world-action / planning models
• Cosmos-H-Dreams interactive simulation
• τ₀-WM unified video-action world modeling
• A2World action-conditioned robotics simulation
• Boundless World Model
• V-JEPA 2 as a predictive representation / world-model backbone

Important taxonomy note

Not every predictive video model is a full world model.

The registry therefore distinguishes between:

• full action-conditioned world models
• world foundation models
• interactive simulators
• planning models
• predictive representation backbones

This is particularly important for V-JEPA 2: the base checkpoint is action-free, while the associated research post-trains V-JEPA 2-AC as an action-conditioned world model for robotic planning.

Intended uses

This dataset can support:

• model discovery
• world-model landscape mapping
• registry interfaces
• comparison tools
• robotics research
• Physical AI research
• planning research
• simulation research
• taxonomy analysis

Limitations

This is an intentionally small first release.

It is not:

• a complete list of all world models
• a benchmark
• a model-quality ranking
• a legal opinion
• a safety certification

Data architecture

WORLD MODELS ORGANIZATION
        ↓
WORLD MODEL REGISTRY DATASET
        ↓
STRUCTURED TAXONOMY + VERIFIED METADATA
        ↓
WORLD MODEL REGISTRY SPACE
        ↓
SEARCH · FILTER · DETAILS · COMPARE

Collaboration & Partnerships

The World Models organization is open to collaboration with world-model research teams, robotics companies, Physical AI companies, simulation platforms, research labs, universities and infrastructure providers.

Relevant collaboration areas include:

• verified registry contributions
• taxonomy
• world-action models
• interactive world models
• Physical AI
• robotics
• planning
• control
• simulation
• benchmarks
• dataset integration

Collaboration: agenten@magenta.de

Independence

World Models is an independent Hugging Face community organization.

This registry is not an official registry of Hugging Face, NVIDIA, Meta, SII, BLM Lab or any other model developer listed here.

License

The original registry structure, curation and documentation are released under Apache-2.0.

This does not relicense any listed model. Each model remains subject to its own license and terms.