# World models for robotics and driving

These action-conditioned models predict what a robot or a vehicle will see after it acts.

Teams use them to test and train control policies in simulation before anything moves in the real world.

- [1X World Model](/world-models/1x-world-model): Robot action and video modeling. Open weights; Permissive license; No real-time claim.
- [comma.ai world model 4B](/world-models/comma-worldmodel-4b): Frame history, ego motion and future plan anchors. Open weights; Permissive license; No real-time claim.
- [Cosmos Predict2.5](/world-models/cosmos-predict2-5): Text, image and video conditioning. Open weights; Custom terms; No real-time claim.
- [Ctrl-World](/world-models/ctrl-world): Keyboard commands converted into robot action chunks. Open weights; Permissive license; No real-time claim.
- [EnerVerse-AC](/world-models/enerverse-ac): Robot actions and camera parameters. Open weights; Non-commercial; No real-time claim.
- [Oasis 3](/world-models/oasis-3): Steering, movement and API control signals. No public weights; Service terms; Real-time, per authors.
- [Odyssey-3](/world-models/odyssey-3): A prompt, then a person's or an agent's actions. No public weights; Service terms; Real-time, per authors.
- [Orbis](/world-models/orbis): Past driving video frames. Open weights; Permissive license; No real-time claim.
- [Orbis 2](/world-models/orbis-2): Past driving video and an optional steering trajectory. Open weights; Permissive license; No real-time claim.
- [Vista](/world-models/vista): Driving trajectory conditioning. Open weights; Permissive license; No real-time claim.

Updated 2026-11-03. Full directory: /world-models