Best AI World Models

NVIDIA Cosmos is a developer-first platform of state-of-the-art generative World Foundation Models (WFMs), advanced video tokenizers, guardrails, and an accelerated data processing and curation pipeline designed to supercharge physical AI development. It enables developers working on autonomous vehicles, robotics, and video analytics AI agents to generate photorealistic, physics-aware synthetic video data, trained on an immense dataset including 20 million hours of real-world and simulated video, to rapidly simulate future scenarios, train world models, and fine‑tune custom behaviors. It includes three core WFM types; Cosmos Predict, capable of generating up to 30 seconds of continuous video from multimodal inputs; Cosmos Transfer, which adapts simulations across environments and lighting for versatile domain augmentation; and Cosmos Reason, a vision-language model that applies structured reasoning to interpret spatial-temporal data for planning and decision-making.

Genie 3 is DeepMind’s next-generation, general-purpose world model capable of generating richly interactive 3D environments in real time at 24 frames per second and 720p resolution that remain consistent for several minutes. Prompted by text input, the system constructs dynamic virtual worlds where users (or embodied agents) can navigate and interact with natural phenomena from multiple perspectives, like first-person or isometric. A standout feature is its emergent long-horizon visual memory: Genie 3 maintains environmental consistency over extended durations, preserving off-screen elements and spatial coherence across revisits. It also supports “promptable world events,” enabling users to modify scenes, such as changing weather or introducing new objects, on the fly. Designed to support embodied agent research, Genie 3 seamlessly integrates with agents like SIMA, facilitating goal-based navigation and complex task accomplishment.

Marble is an experimental AI model internally tested by World Labs, a variant and extension of their Large World Model technology. It is a web service that turns a single 2D image into a navigable spatial environment. Marble offers two generation modes: a smaller, fast model for rough previews that’s quick to iterate on, and a larger, high-fidelity model that takes longer (around ten minutes in the example) but produces a significantly more convincing result. The value proposition is instant, photogrammetry-like image-to-world creation without a full capture rig, turning a single shot into an explorable space for memory capture, mood boards, archviz previews, or creative experiments.

Mirage 2 is an AI-driven Generative World Engine that lets anyone instantly transform images or descriptions into fully playable, interactive game environments directly in the browser. Upload sketches, concept art, photos, or prompts, like “Ghibli-style village” or “Paris street scene”, and Mirage 2 builds immersive worlds you can explore in real time. The experience isn’t pre-scripted: you can modify your world mid-play using natural-language chat, evolving settings dynamically, from a cyberpunk city to a rainforest or a mountaintop castle, all with minimal latency (around 200 ms) on a single consumer GPU. Mirage 2 supports smooth rendering, real-time prompt control, and extended gameplay stretches beyond ten minutes. It outpaces earlier world-model systems by offering true general-domain generation, no upper limit on styles or genres, as well as seamless world adaptation and sharing features.

Odyssey is a frontier interactive video model that enables instant, real-time generation of video you can interact with. Just type a prompt, and the system begins streaming minutes of video that respond to your input. It shifts video from a static playback format to a dynamic, action-aware stream: the model is causal and autoregressive, generating each frame based solely on prior frames and your actions rather than a fixed timeline, enabling continuous adaptation of camera angles, scenery, characters, and events. The platform begins streaming video almost instantly, producing new frames every ~50 milliseconds (about 20 fps), so you don’t wait minutes for a clip, you engage in an evolving experience. Under the hood, the model is trained via a novel multi-stage pipeline to transition from fixed-clip generation to open-ended interactive video, allowing you to type or speak commands and explore an AI-imagined world that reacts in real time.

GWM-1 is Runway’s state-of-the-art General World Model designed to simulate the real world in real time. It is an interactive, controllable, and general-purpose model built on top of Runway’s Gen-4.5 architecture. GWM-1 generates high-fidelity video frame by frame while maintaining long-term spatial and behavioral consistency. The model supports action-conditioning through inputs such as camera movement, robot actions, events, and speech. GWM-1 enables realistic visual simulation paired with synchronized video and audio outputs. It is designed to help AI systems experience environments rather than just describe them. GWM-1 represents a major step toward general-purpose simulation beyond language-only models.

Active Inference is a novel framework for agentic AI based on world models, emerging from over 30 years of research in computational neuroscience. From this paradigm, we offer an AI built for power and computational efficiency, designed to live on-device and on the edge. Integrating with traditional computer vision stacks our intelligent decision-making systems provide an explainable output that allows organizations to build accountability into their AI tools and products. We are taking active inference from neuroscience into AI as the foundation for software that will allow robots and embodied platforms to make autonomous decisions like the human brain.

Game Worlds is an emerging AI-powered gaming platform developed by Runway, a company known for pioneering generative AI tools in Hollywood. This new platform aims to let users create and explore video games generated with AI technology, simplifying game development. Currently, Game Worlds features a chat interface that supports text and image generation, with full AI-generated video games planned for release later in 2025. Runway’s CEO envisions AI accelerating game development much like it has in film production, making game creation faster and more accessible. The platform is positioned as a breakthrough for gamers and developers seeking innovative ways to build and interact with games. Game Worlds represents the future of AI-driven game design and interactive experiences.