https://world-models.io The latest AI world models, research, timeline events, and guides for model-based RL, robotics, and embodied AI. en Wed, 29 Jul 2026 16:01:01 GMT https://world-models.io/en/timeline/world-labs-blog-building-worlds-that-train-robots-1785265653541 When spatial intelligence becomes physical, the north star goal is to advance the field of robotics. Tue, 28 Jul 2026 00:00:00 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-building-worlds-that-train-robots-1785265653541 https://world-models.io/en/timeline/hugging-face-blog-nvidia-cosmos-h-dreams-bringing-real-time-generative-simulatio Today, we are introducing the next step: Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics. Mon, 27 Jul 2026 09:32:20 GMT Lab Update https://world-models.io/en/timeline/hugging-face-blog-nvidia-cosmos-h-dreams-bringing-real-time-generative-simulatio https://world-models.io/en/models/dreamer-v3 A general algorithm for mastering diverse domains with fixed hyperparameters through world model learning. Fri, 24 Jul 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/dreamer-v3 https://world-models.io/en/timeline/hugging-face-blog-introducing-cosmos-3-edge-1784567224835 Today we are releasing NVIDIA Cosmos 3 Edge on Hugging Face's Cosmos 3 repository. It's a 4-billion-parameter open world model that helps robots and vision AI agents understand their surroundings, reason in real time and generate robot actions on edge devices. Mon, 20 Jul 2026 15:58:51 GMT Lab Update https://world-models.io/en/timeline/hugging-face-blog-introducing-cosmos-3-edge-1784567224835 https://world-models.io/en/timeline/cohere-research-introducing-north-mini-code-cohere-s-first-model-for-developers- North Mini Code's benchmark scores translate to a 33. 4 on the Artificial Analysis Coding Index, a competitive position among similarly sized models. North Mini Code is designed for speed and efficiency, with a strong focus on minimizing total cost of ownership as we continue to refine and scale the model. Sun, 19 Jul 2026 16:07:50 GMT Lab Update https://world-models.io/en/timeline/cohere-research-introducing-north-mini-code-cohere-s-first-model-for-developers- https://world-models.io/en/models/wham Microsoft Research's World and Human Action Model, explicitly designed to model game environments together with human actions. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/wham https://world-models.io/en/models/wham-rt The real-time member of Microsoft's WHAM family of World and Human Action Models. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/wham-rt https://world-models.io/en/models/gaia-2 Wayve's controllable multi-view world model for generating driving scenarios from structured conditioning signals. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/gaia-2 https://world-models.io/en/models/waabi-world Waabi's generative, closed-loop world simulator for designing, assessing, and training autonomous driving systems. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/waabi-world https://world-models.io/en/models/odyssey-2 Odyssey's general-purpose world model for real-time, open-ended interactive video. Sun, 19 Jul 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/odyssey-2 https://world-models.io/en/models/hy-world-2 Tencent Hunyuan's open multimodal world model for reconstructing, generating, and simulating navigable 3D worlds. Sun, 19 Jul 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/hy-world-2 https://world-models.io/en/models/relic Adobe Research's interactive video world model combining real-time exploration, long-horizon spatial memory, and user control. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/relic https://world-models.io/en/models/matrix-game-2 Skywork AI's open-source, real-time streaming interactive world model for action-conditioned video generation. Sun, 19 Jul 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/matrix-game-2 https://world-models.io/en/timeline/waabi-insights-waabi-proves-autonomous-truck-generalization-with-volvo-autonomou The self-driving company announced in a blog post that its Waabi Driver software, trained exclusively on a Peterbilt 579, took control of a Volvo VNL Autonomous truck and drove it safely on highways and complex surface streets from the very first mile. No new real-world data. No simulation data. Wed, 15 Jul 2026 14:47:58 GMT Lab Update https://world-models.io/en/timeline/waabi-insights-waabi-proves-autonomous-truck-generalization-with-volvo-autonomou https://world-models.io/articles/building-a-practical-taxonomy-for-ai-world-models Why comparing world models is harder than it looks, and how a practical taxonomy can make the field easier to understand. Tue, 14 Jul 2026 00:00:00 GMT Blog https://world-models.io/articles/building-a-practical-taxonomy-for-ai-world-models https://world-models.io/en/research/video-world-models How video world models learn physics, temporal consistency, and interactive simulation from large-scale video, from Sora and Genie to Cosmos and V-JEPA. Sat, 20 Jun 2026 00:00:00 GMT Research https://world-models.io/en/research/video-world-models https://world-models.io/en/research/diffusion-world-models How diffusion world models generate future states, preserve richer visual detail, and power video simulation systems such as DIAMOND, Sora, and Cosmos. Sat, 20 Jun 2026 00:00:00 GMT Research https://world-models.io/en/research/diffusion-world-models https://world-models.io/en/research/world-model-evaluation How to evaluate world models across rollout quality, benchmark performance, planning utility, and downstream transfer instead of relying on visual plausibility alone. Sat, 20 Jun 2026 00:00:00 GMT Research https://world-models.io/en/research/world-model-evaluation https://world-models.io/en/research/language-conditioned-world-models How language-conditioned world models use text prompts or natural-language actions to control simulation, planning, and embodied behavior across Pandora, 3D-VLA, RT-2, and hybrid systems. Sat, 20 Jun 2026 00:00:00 GMT Research https://world-models.io/en/research/language-conditioned-world-models https://world-models.io/en/compare/playworld-vs-td-mpc2 Two green-index models for robot decision-making, but with very different operating modes. PlayWorld learns a manipulation-focused world simulator from autonomous play, while TD-MPC2 combines latent dynamics with model-predictive control across a wide multi-task control benchmark suite. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/playworld-vs-td-mpc2 https://world-models.io/en/compare/genie-3-vs-nvidia-cosmos Two green-index frontier systems with different ambitions. Genie 3 is a real-time text-to-world interactive generator, while NVIDIA Cosmos is a broad physical-AI platform optimized for simulation infrastructure, robotics, and industrial world modeling. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/genie-3-vs-nvidia-cosmos https://world-models.io/en/compare/genie-3-vs-v-jepa-2 Two green-index leaders that represent different frontier philosophies. Genie 3 is an interactive generative world model that turns text into playable environments, while V-JEPA 2 is a self-supervised latent predictor optimized for physical reasoning and zero-shot robot planning. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/genie-3-vs-v-jepa-2 https://world-models.io/en/compare/nvidia-cosmos-vs-v-jepa-2 Two green-index foundation-scale leaders with different views of world modeling. Cosmos emphasizes a platform for physical-AI simulation and generation, while V-JEPA 2 emphasizes self-supervised predictive representations for visual understanding and robot control. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/nvidia-cosmos-vs-v-jepa-2 https://world-models.io/en/compare/playworld-vs-v-jepa-2 Two green-index models pushing robotics-relevant world understanding in different ways. PlayWorld is a robot-play simulator for manipulation and policy improvement, while V-JEPA 2 is a self-supervised video predictor optimized for physical reasoning and zero-shot robot planning. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/playworld-vs-v-jepa-2 https://world-models.io/en/compare/dreamer-v3-vs-playworld Two green-index leaders for acting under learned dynamics, but with different centers of gravity. DreamerV3 is the canonical imagination-based general RL agent, while PlayWorld is a manipulation-centric robot simulator trained from autonomous play data. Thu, 18 Jun 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-playworld https://world-models.io/en/models/marble World Labs' multimodal world model for generating spatially consistent, persistent 3D environments from text, images, video, and 360 inputs. Fri, 12 Jun 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/marble https://world-models.io/en/models/1x-world-model 1X's physics-grounded video world model for anticipating the outcomes of NEO's actions and supporting generalization to unseen household tasks. Fri, 12 Jun 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/1x-world-model https://world-models.io/en/models/playworld A Princeton robot world model trained from autonomous self-play to simulate contact-rich manipulation and support policy evaluation and RL fine-tuning. Fri, 12 Jun 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/playworld https://world-models.io/en/timeline/arxiv-cs-ai-business-world-model-1781064567990 Businesses are increasingly adopting AI-enabled tools to improve productivity, reduce costs, and enhance products and services. Wed, 10 Jun 2026 04:00:00 GMT Paper Published https://world-models.io/en/timeline/arxiv-cs-ai-business-world-model-1781064567990 https://world-models.io/en/timeline/world-labs-blog-taxonomy-of-world-models-1780614566917 The World Is Not Made of Words In an earlier essay, we argued that spatial intelligence is AI's next frontier and that world models are the path to it. Thu, 04 Jun 2026 23:09:27 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-taxonomy-of-world-models-1780614566917 https://world-models.io/en/timeline/arxiv-cs-lg-world-models-a-comprehensive-survey-of-architectures-methodologies-r World models, internal simulators that learn the structure and dynamics of an environment, have emerged as a central paradigm in the pursuit of artificial general intelligence, enabling agents to predict, plan, and reason within learned representations. Tue, 02 Jun 2026 04:00:00 GMT Paper Published https://world-models.io/en/timeline/arxiv-cs-lg-world-models-a-comprehensive-survey-of-architectures-methodologies-r https://world-models.io/en/timeline/hugging-face-blog-welcome-nvidia-cosmos-3-the-first-open-omni-model-for-physical Cosmos 3 represents a major leap forward in world foundation models (WFMs) for physical AI: a single, unified omni-model that combines world generation, physical reasoning, and action generation in one model. Mon, 01 Jun 2026 04:44:55 GMT Lab Update https://world-models.io/en/timeline/hugging-face-blog-welcome-nvidia-cosmos-3-the-first-open-omni-model-for-physical https://world-models.io/en/timeline/arxiv-cs-ai-physically-viable-world-models-a-case-for-query-conditioned-embodied World models for embodied AI must be physically viable: constructed to answer intervention queries by representing the physical structure governing action outcomes, rather than merely predicting future observations. Mon, 01 Jun 2026 04:00:00 GMT Paper Published https://world-models.io/en/timeline/arxiv-cs-ai-physically-viable-world-models-a-case-for-query-conditioned-embodied https://world-models.io/en/timeline/arxiv-cs-lg-emergent-semantic-representations-in-world-models-through-physical-i What does a world model learn from physical exploration, without any linguistic supervision? We argue the answer is organized by a single principle: the geometric structure of the physical world. Fri, 29 May 2026 04:00:00 GMT Paper Published https://world-models.io/en/timeline/arxiv-cs-lg-emergent-semantic-representations-in-world-models-through-physical-i https://world-models.io/en/timeline/google-deepmind-blog-catch-up-on-the-dialogues-stage-at-google-i-o-2026-17794732 A recap of the 2026 I/O Dialogues, where leaders discuss the future of AI, quantum computing, robotics and creativity. Fri, 22 May 2026 18:00:00 GMT Lab Update https://world-models.io/en/timeline/google-deepmind-blog-catch-up-on-the-dialogues-stage-at-google-i-o-2026-17794732 https://world-models.io/en/timeline/arxiv-cs-lg-prowl-prioritized-regret-driven-optimization-for-world-model-learnin Modern action-conditioned video world models achieve strong short-horizon visual realism, yet remain unreliable on rare, interaction-critical transitions that dominate downstream planning and policy performance. Wed, 20 May 2026 04:00:00 GMT Paper Published https://world-models.io/en/timeline/arxiv-cs-lg-prowl-prioritized-regret-driven-optimization-for-world-model-learnin https://world-models.io/en/timeline/hugging-face-blog-fine-tuning-nvidia-cosmos-predict-2-5-with-lora-dora-for-robot This article describes fine-tuning NVIDIA Cosmos Predict 2. 5, a 2B-parameter world model for generating physically plausible videos, using LoRA and DoRA. Full fine-tuning is resource-intensive and risks catastrophic forgetting. Mon, 18 May 2026 16:00:21 GMT Lab Update https://world-models.io/en/timeline/hugging-face-blog-fine-tuning-nvidia-cosmos-predict-2-5-with-lora-dora-for-robot https://world-models.io/en/compare/sora-vs-genie-3 Two leading generative world models with opposite design goals: Sora targets long, cinematic, non-interactive clips from text, while Genie 3 trades fidelity for real-time controllable worlds you can actually play. Tue, 12 May 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/sora-vs-genie-3 https://world-models.io/en/compare/pixverse-r1-vs-sora PixVerse R1 introduces reasoning-trained generation to text-to-video, optimizing for prompt adherence and physical plausibility. Sora remains the reference for cinematic length and visual fidelity. Tue, 12 May 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/pixverse-r1-vs-sora https://world-models.io/en/compare/v-jepa-2-vs-i-jepa Two milestones of the JEPA roadmap: I-JEPA established self-supervised image representation by predicting in latent space; V-JEPA 2 extends the paradigm to video at foundation scale and demonstrates zero-shot robot control. Tue, 12 May 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/v-jepa-2-vs-i-jepa https://world-models.io/en/compare/pandora-vs-genie-2 Both generate explorable 3D-feeling environments, but with different control surfaces: Pandora accepts free-form text actions through an LLM backbone, while Genie 2 conditions on a single seed image and learned latent actions. Tue, 12 May 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/pandora-vs-genie-2 https://world-models.io/en/compare/leworldmodel-vs-dreamer-v3 LeWorldModel revisits LeCun's energy-based JEPA philosophy for control, predicting in latent space without pixel reconstruction. DreamerV3 remains the canonical RSSM-based agent that learns by imagining pixel-grounded rollouts. Tue, 12 May 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/leworldmodel-vs-dreamer-v3 https://world-models.io/en/timeline/world-labs-blog-spark-2-0-1777371384042 Spark 2. 0 introduces a Level-of-Detail (LoD) system for streaming large 3D Gaussian Splatting (3DGS) worlds on the web. Built on THREE. js and WebGL2, Spark renders complex 3DGS scenes across various devices including desktop, iOS, Android, and VR. Tue, 28 Apr 2026 10:16:24 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-spark-2-0-1777371384042 https://world-models.io/en/timeline/runway-research-gen-4-5-the-world-s-best-video-model-featuring-state-of-the-art- Runway Gen-4. 5 is a leading AI video generation model, offering advanced visual fidelity and creative control. Built on NVIDIA Hopper and Blackwell GPUs, it produces cinematic, realistic, and highly controllable video outputs. Wed, 15 Apr 2026 16:45:31 GMT Model Release https://world-models.io/en/timeline/runway-research-gen-4-5-the-world-s-best-video-model-featuring-state-of-the-art- https://world-models.io/en/models/genie-3 Google DeepMind's general-purpose world model that generates interactive 3D environments from text prompts in real time at 24fps. Fri, 10 Apr 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/genie-3 https://world-models.io/en/models/v-jepa-2 Meta FAIR's self-supervised video world model achieving state-of-the-art visual understanding and enabling zero-shot robot control. Fri, 10 Apr 2026 00:00:00 GMT Self-Supervised World Model https://world-models.io/en/models/v-jepa-2 https://world-models.io/en/models/leworldmodel A compact 15M-parameter JEPA world model that learns real-world physics on a single GPU, solving the notorious representation collapse problem. Fri, 10 Apr 2026 00:00:00 GMT Self-Supervised World Model https://world-models.io/en/models/leworldmodel https://world-models.io/en/models/pixverse-r1 The first real-time world model supporting multi-user shared worlds with personalized avatars and no session limits. Fri, 10 Apr 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/pixverse-r1 https://world-models.io/en/compare/genie-3-vs-genie-2 Two generations of DeepMind's interactive world model. Genie 2 generates 3D environments from single images; Genie 3 generates them from text prompts in real time at 24fps with far greater diversity and consistency. Fri, 10 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/genie-3-vs-genie-2 https://world-models.io/en/compare/v-jepa-2-vs-v-jepa V-JEPA 2 dramatically scales up Meta FAIR's self-supervised video world model, achieving state-of-the-art visual understanding and zero-shot robot control, capabilities V-JEPA didn't demonstrate. Fri, 10 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/v-jepa-2-vs-v-jepa https://world-models.io/en/models/rt-2 A vision-language-action model that transfers web-scale knowledge directly to robot control. Tue, 07 Apr 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/rt-2 https://world-models.io/en/models/lwm A foundation model trained on 1M+ interleaved video and language tokens for long-horizon world understanding. Tue, 07 Apr 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/lwm https://world-models.io/en/models/stable-video-diffusion An open-source foundation model for video generation, producing temporally consistent video from a single image. Tue, 07 Apr 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/stable-video-diffusion https://world-models.io/en/models/mile A world model for autonomous driving that jointly learns dynamics, perception, and planning through model-based imitation learning. Tue, 07 Apr 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/mile https://world-models.io/en/models/steve-1 An instruction-following agent for Minecraft that uses a generative world model to execute open-ended text commands. Tue, 07 Apr 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/steve-1 https://world-models.io/en/models/gen-3-alpha Runway's next-generation video model with fine-grained control over motion, style, and composition. Tue, 07 Apr 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/gen-3-alpha https://world-models.io/en/compare/rt-2-vs-3d-vla Two approaches to vision-language-action models for robotics. RT-2 leverages web-scale VLM knowledge through action tokenization, while 3D-VLA integrates explicit 3D spatial understanding for embodied reasoning. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/rt-2-vs-3d-vla https://world-models.io/en/compare/sora-vs-gen-3-alpha The two leading commercial video generation models. Sora emphasizes physical world simulation and long-form coherence, while Gen-3 Alpha focuses on fine-grained creative control and production-ready tools. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/sora-vs-gen-3-alpha https://world-models.io/en/compare/stable-video-diffusion-vs-emu-video Two image-to-video models: SVD is open-source and community-driven, while Emu Video is Meta's factorized approach that separates image and motion generation for better controllability. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/stable-video-diffusion-vs-emu-video https://world-models.io/en/compare/mile-vs-gaia-1 Both from Wayve, these models represent two generations of driving world models. MILE focuses on actionable imagination for planning, while GAIA-1 scales to photorealistic scenario generation. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/mile-vs-gaia-1 https://world-models.io/en/compare/steve-1-vs-dreamer-v3 Two approaches to open-world game AI. STEVE-1 uses video pre-training and instruction following, while DreamerV3 learns a world model from scratch via reinforcement learning. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/steve-1-vs-dreamer-v3 https://world-models.io/en/compare/lwm-vs-v-jepa Two approaches to learning world understanding from video. LWM uses autoregressive prediction over million-length sequences, while V-JEPA predicts abstract latent representations without pixel reconstruction. Tue, 07 Apr 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/lwm-vs-v-jepa https://world-models.io/en/timeline/dreamerv3-nature-publication Dreamer, third generation, is a general reinforcement learning algorithm that outperforms specialized methods across over 150 diverse control tasks with a single configuration. Fri, 03 Apr 2026 10:00:00 GMT Paper Published https://world-models.io/en/timeline/dreamerv3-nature-publication https://world-models.io/en/timeline/wm-bench-benchmark WM Bench is a new benchmark assessing cognitive intelligence in world models, available on Hugging Face. It measures comprehension beyond realistic rendering or natural flow, focusing on a model's ability to "think" and understand scenarios. Sun, 29 Mar 2026 11:00:00 GMT Benchmark Update https://world-models.io/en/timeline/wm-bench-benchmark https://world-models.io/en/timeline/genie3-gdc-talk-consistency Google DeepMind presented Genie 3, a generative AI technology, at GDC This world model consistently generates environments for about one minute before inconsistencies appear, a significant improvement from earlier versions. Wed, 25 Mar 2026 18:00:00 GMT Lab Update https://world-models.io/en/timeline/genie3-gdc-talk-consistency https://world-models.io/en/timeline/meta-vjepa2-official-release V-JEPA 2 is an open-source world model developed by Meta FAIR, demonstrating state-of-the-art performance in visual understanding and prediction for the physical world. This 1. Wed, 25 Mar 2026 09:00:00 GMT Model Release https://world-models.io/en/timeline/meta-vjepa2-official-release https://world-models.io/en/timeline/deepmind-gemini-robotics Google DeepMind introduces Gemini Robotics and Gemini Robotics-ER, two AI models built on Gemini 2. 0, designed to bring advanced AI capabilities to the physical world through robotics. Sun, 22 Mar 2026 14:00:00 GMT Model Release https://world-models.io/en/timeline/deepmind-gemini-robotics https://world-models.io/en/timeline/arya-physics-world-model ARYA is a physics-constrained, composable, and deterministic world model architecture founded on five principles: nano models, composability, causal reasoning, determinism, and architectural AI safety. Sun, 22 Mar 2026 10:00:00 GMT Paper Published https://world-models.io/en/timeline/arya-physics-world-model https://world-models.io/en/compare/dreamer-v2-vs-dreamer-v3 The Dreamer lineage's two most impactful iterations: DreamerV2 achieved human-level Atari with discrete representations, while DreamerV3 eliminated hyperparameter tuning entirely with symlog predictions. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v2-vs-dreamer-v3 https://world-models.io/en/compare/sora-vs-emu-video Two generative video models from competing labs: Sora represents OpenAI's vision of video as world simulation, while Emu Video is Meta's efficient factorized approach to high-quality text-to-video generation. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/sora-vs-emu-video https://world-models.io/en/compare/iris-vs-diamond Two approaches to learning game simulators: IRIS uses discrete tokenization with a GPT-like transformer, while DIAMOND leverages diffusion models for higher visual fidelity. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/iris-vs-diamond https://world-models.io/en/compare/v-jepa-vs-nvidia-cosmos Two foundation-scale approaches to world understanding: V-JEPA learns predictive video representations through self-supervised masking, while Cosmos builds a full-stack world simulation platform for physical AI. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/v-jepa-vs-nvidia-cosmos https://world-models.io/en/compare/ami-vs-ha-world-model Two pioneering cognitive-inspired world models: Ha's 2018 World Model introduced the VAE+RNN+Controller architecture, while AMI proposes an autonomous machine intelligence framework inspired by biological cognition. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/ami-vs-ha-world-model https://world-models.io/en/compare/gaia-1-vs-sora Two generative world models that approach video generation from different angles: GAIA-1 focuses on autonomous driving simulation, while Sora aims to be a general-purpose visual world simulator. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/gaia-1-vs-sora https://world-models.io/en/compare/oasis-vs-pandora Two real-time neural game engines: OASIS generates Minecraft-like worlds at 20+ FPS using latent diffusion, while Pandora creates diverse game worlds using a hybrid autoregressive-diffusion architecture. Sat, 21 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/oasis-vs-pandora https://world-models.io/en/timeline/leworldmodel-compact-jepa LeWorldModel (LeWM) is introduced as the first Joint Embedding Predictive Architecture (JEPA) that achieves stable end-to-end training from raw pixels. It employs only two loss terms, reducing hyperparameter tuning. Fri, 20 Mar 2026 10:00:00 GMT Paper Published https://world-models.io/en/timeline/leworldmodel-compact-jepa https://world-models.io/en/compare/dreamer-v2-vs-planet Both use the RSSM architecture for latent dynamics, but DreamerV2 introduced discrete representations that dramatically improved performance. PlaNet pioneered the approach; DreamerV2 perfected it for Atari-scale environments. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v2-vs-planet https://world-models.io/en/compare/dreamer-v3-vs-td-mpc2 Two leading model-based RL agents with different philosophies: DreamerV3 uses imagination-based actor-critic learning, while TD-MPC2 combines temporal-difference learning with model-predictive control for multi-task mastery. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-td-mpc2 https://world-models.io/en/compare/genie-2-vs-unisim Both are generative world models that create interactive environments, but Genie 2 generates 3D worlds from single images while UniSim learns a universal action-conditioned simulator from diverse real-world data. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/genie-2-vs-unisim https://world-models.io/en/compare/muzero-vs-td-mpc2 Both use learned dynamics models for planning, but MuZero uses Monte Carlo tree search for deep discrete planning while TD-MPC2 uses model-predictive control for continuous multi-task settings. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/muzero-vs-td-mpc2 https://world-models.io/en/compare/nvidia-cosmos-vs-genie-2 Two foundation-scale world models with different strategies: Cosmos is an open industrial platform for physical AI training, while Genie 2 is a DeepMind research system that generates interactive 3D environments from images. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/nvidia-cosmos-vs-genie-2 https://world-models.io/en/compare/gaia-1-vs-nvidia-cosmos Both are video-based world models for autonomous driving and physical AI, but GAIA-1 is a domain-specific driving world model from Wayve while Cosmos is a general-purpose foundation platform from NVIDIA. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/gaia-1-vs-nvidia-cosmos https://world-models.io/en/compare/v-jepa-vs-i-jepa Both implement Yann LeCun's JEPA framework for self-supervised learning, but V-JEPA operates on video (temporal dynamics) while I-JEPA operates on static images (spatial structure). Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/v-jepa-vs-i-jepa https://world-models.io/en/compare/oasis-vs-diamond Both use diffusion models as world models for interactive environments, but OASIS generates real-time playable Minecraft-like worlds while DIAMOND uses diffusion for model-based RL training in Atari. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/oasis-vs-diamond https://world-models.io/en/compare/genie-1-vs-genie-2 Genie pioneered unsupervised interactive environment generation from video. Genie 2 massively scales this approach to generate persistent, interactive 3D worlds from single images. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/genie-1-vs-genie-2 https://world-models.io/en/compare/dreamer-v3-vs-iris Two model-based RL agents using fundamentally different world model architectures: DreamerV3's RSSM with actor-critic vs. IRIS's autoregressive Transformer with VQ-VAE tokens. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-iris https://world-models.io/en/compare/sora-vs-nvidia-cosmos Both generate video from learned world dynamics, but Sora is a creative video generation model while Cosmos is an industrial platform for physical AI training and simulation. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/sora-vs-nvidia-cosmos https://world-models.io/en/compare/ha-world-model-vs-dreamer-v3 The original 2018 'World Models' paper vs. the current state-of-the-art: how five years of research transformed a foundational concept into a domain-general world model agent. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/ha-world-model-vs-dreamer-v3 https://world-models.io/en/compare/muzero-vs-dreamer-v3 Two titans of model-based RL with fundamentally different approaches: MuZero learns a value-equivalent model for search-based planning, while DreamerV3 learns a generative world model for imagination-based policy optimization. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/muzero-vs-dreamer-v3 https://world-models.io/en/compare/pandora-vs-oasis Both generate interactive game-like worlds, but Pandora produces multi-domain video simulations with narrative control, while OASIS focuses on high-fidelity real-time open-world generation trained on Minecraft. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/pandora-vs-oasis https://world-models.io/en/compare/copilot4d-vs-gaia-1 Both target autonomous driving simulation but from different angles: Copilot4D predicts 4D point cloud futures for safety-critical planning, while GAIA-1 generates photorealistic driving video for scenario exploration. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/copilot4d-vs-gaia-1 https://world-models.io/en/compare/emu-video-vs-sora Both are frontier video generation models, but with different ambitions: Emu Video focuses on efficient, high-quality short-form generation, while Sora pushes toward long-form, physically coherent world simulation. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/emu-video-vs-sora https://world-models.io/en/compare/3d-vla-vs-i-jepa Two approaches to learning representations for embodied intelligence: 3D-VLA combines 3D perception with language-conditioned action planning, while I-JEPA learns abstract visual representations through self-supervised prediction in latent space. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/3d-vla-vs-i-jepa https://world-models.io/en/compare/gamengen-vs-diamond Both simulate game environments in real-time, but with radically different approaches: GameNGen uses a fine-tuned diffusion model for photorealistic DOOM simulation, while DIAMOND uses a diffusion-based world model for Atari with reinforcement learning. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/gamengen-vs-diamond https://world-models.io/en/compare/predictron-vs-muzero Both learn abstract dynamics models for planning without requiring environment reconstruction, but Predictron was an early prototype while MuZero became the definitive realization of value-equivalent model learning. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/predictron-vs-muzero https://world-models.io/en/compare/unisim-vs-genie-2 Both are large-scale generative world simulators, but UniSim focuses on unified simulation across real-world domains while Genie 2 generates persistent, explorable 3D environments from single images. Fri, 20 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/unisim-vs-genie-2 https://world-models.io/en/timeline/nvidia-cosmos-gtc-2025-update NVIDIA introduced new open-source tools for physical AI development at GTC Key releases include Cosmos Transfer, a 7-billion-parameter world foundation model for generating virtual scenes with multicontrols, and the open Physical AI Dataset, a 15TB commercial-grade dataset for robotics training. Wed, 18 Mar 2026 15:30:00 GMT Model Release https://world-models.io/en/timeline/nvidia-cosmos-gtc-2025-update https://world-models.io/en/models/sora OpenAI's video generation model that simulates the physical world by generating realistic videos from text prompts. Wed, 18 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/sora https://world-models.io/en/compare/sora-vs-genie-2 Sora and Genie 2 both generate video from prompts, but they approach world simulation very differently. Sora generates passive, high-fidelity videos from text; Genie 2 generates interactive, controllable 3D environments from images. Wed, 18 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/sora-vs-genie-2 https://world-models.io/en/compare/oasis-vs-gamengen OASIS and GameNGen both demonstrate neural networks functioning as real-time game engines, but they target different games and use different architectures. They represent the emerging frontier of neural game engines. Tue, 17 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/oasis-vs-gamengen https://world-models.io/en/timeline/world-labs-blog-bigger-better-worlds-1773702003953 World Labs introduces Marble, a limited-access beta for generating larger, more detailed 3D worlds from image or text prompts. This new model produces persistent, navigable, and controllable environments with improved fidelity, stylistic diversity, and cleaner geometry compared to previous iterations. Mon, 16 Mar 2026 23:00:03 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-bigger-better-worlds-1773702003953 https://world-models.io/en/timeline/world-labs-blog-generating-worlds-1773702004014 This collection highlights progress in generating persistent, navigable 3D worlds accessible through a web browser. It features early developments, a technical deep dive into Spark 2. Mon, 16 Mar 2026 23:00:03 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-generating-worlds-1773702004014 https://world-models.io/en/timeline/world-labs-blog-announcing-the-world-api-1773692071407 World Labs introduces the World API, enabling developers to generate explorable 3D worlds from various inputs like text, images, panoramas, and videos using their multimodal world model, Marble. Mon, 16 Mar 2026 20:14:23 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-announcing-the-world-api-1773692071407 https://world-models.io/en/timeline/world-labs-blog-3d-as-code-1773692063661 This article posits 3D as the universal interface for spatial interaction, akin to text for software. It argues that 3D representations function as a "spatial code," facilitating communication and collaboration between humans and AI, and between machines themselves. Mon, 16 Mar 2026 20:14:23 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-3d-as-code-1773692063661 https://world-models.io/en/timeline/wayve-blog-research-1773692060379 Wayve's research archives highlight advancements in autonomous driving using embodied AI. Key projects include LA-Pose for camera pose estimation from vast unlabeled video data, and Rig3R for 3D perception and ego-motion. Mon, 16 Mar 2026 20:14:20 GMT Model Release https://world-models.io/en/timeline/wayve-blog-research-1773692060379 https://world-models.io/en/timeline/lecun-ami-labs-1b-funding Yann LeCun's new venture, AMI Labs, secured a record $1. 03 billion seed round to develop "world models" as an alternative to large language models (LLMs). Mon, 16 Mar 2026 10:00:00 GMT Lab Update https://world-models.io/en/timeline/lecun-ami-labs-1b-funding https://world-models.io/en/models/ami-world-model A multimodal world foundation model designed for embodied AI, combining visual, proprioceptive, and language understanding for robot learning. Mon, 16 Mar 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/ami-world-model https://world-models.io/en/models/3d-vla A 3D vision-language-action model with a built-in world model for embodied AI, enabling 3D-aware reasoning, planning, and action generation. Mon, 16 Mar 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/3d-vla https://world-models.io/en/compare/v-jepa-vs-video-generation V-JEPA and video generation models like Sora both learn from video, but follow opposite philosophies: V-JEPA predicts in abstract representation space without generating pixels, while video generation models focus on producing realistic pixel outputs. Mon, 16 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/v-jepa-vs-video-generation https://world-models.io/en/models/pandora A general world model combining autoregressive and diffusion architectures for generating interactive, controllable video environments. Sun, 15 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/pandora https://world-models.io/en/research/world-models-survey A survey of AI world models covering taxonomy, leading architectures, landmark systems, open challenges, and future research directions. Sun, 15 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/world-models-survey https://world-models.io/en/compare/dreamer-v3-vs-diamond DreamerV3 and DIAMOND are both model-based RL agents that train policies via imagination, but they use fundamentally different dynamics models: RSSM latent dynamics vs. pixel-space diffusion models. Sun, 15 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-diamond https://world-models.io/en/timeline/deepmind-genie-2-release Google DeepMind introduces Genie 2, a foundational world model for generating diverse, interactive 3D environments. Building on its predecessor, Genie 2 can create playable worlds from a single image prompt, allowing for human or AI agent interaction via keyboard and mouse inputs. Sat, 14 Mar 2026 14:30:00 GMT Model Release https://world-models.io/en/timeline/deepmind-genie-2-release https://world-models.io/en/models/genie-2 A foundation world model that generates diverse, playable 3D environments from a single image prompt. Sat, 14 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/genie-2 https://world-models.io/en/models/genie-1 The first generative interactive environment trained from unlabeled internet videos, capable of generating action-controllable 2D worlds. Sat, 14 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/genie-1 https://world-models.io/en/research/foundation-world-models How foundation world models such as Cosmos and Genie 2 bring large-scale learned simulation to robotics, autonomous driving, and physical AI. Sat, 14 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/foundation-world-models https://world-models.io/en/compare/nvidia-cosmos-vs-dreamer Cosmos and DreamerV3 represent two different scales and approaches to world modeling: Cosmos is a foundation-scale video world model platform for physical AI, while DreamerV3 is a sample-efficient RL agent with learned dynamics. Sat, 14 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/nvidia-cosmos-vs-dreamer https://world-models.io/en/compare/gaia-1-vs-copilot4d Both are world models designed for autonomous driving, but they operate on different sensor modalities: GAIA-1 generates camera video, while Copilot4D predicts LiDAR point clouds in 4D. Sat, 14 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/gaia-1-vs-copilot4d https://world-models.io/en/models/v-jepa Video Joint Embedding Predictive Architecture: learns visual world models through self-supervised video prediction in abstract representation space. Fri, 13 Mar 2026 00:00:00 GMT Self-Supervised World Model https://world-models.io/en/models/v-jepa https://world-models.io/en/models/nvidia-cosmos A platform of state-of-the-art generative world foundation models for physical AI development. Thu, 12 Mar 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/nvidia-cosmos https://world-models.io/en/models/oasis An open-source real-time interactive world model that generates playable game environments at 20+ FPS entirely from a neural network. Thu, 12 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/oasis https://world-models.io/en/research/world-models-robotics How world models improve robot learning, learned simulation, safe exploration, and sim-to-real transfer across manipulation, navigation, and control. Thu, 12 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/world-models-robotics https://world-models.io/en/compare/model-based-rl-vs-model-free-rl Model-based RL learns a world model for imagination-based planning. Model-free RL learns directly from interaction without an internal model. Each approach has distinct strengths depending on the application domain. Thu, 12 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/model-based-rl-vs-model-free-rl https://world-models.io/en/compare/iris-vs-dreamer-v3 IRIS and DreamerV3 are both leading model-based RL agents but use fundamentally different world model architectures: autoregressive token prediction vs. RSSM latent dynamics. Thu, 12 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/iris-vs-dreamer-v3 https://world-models.io/en/timeline/dreamerv3-atari-benchmark-update DreamerV3, an artificial intelligence agent, has achieved human-level performance on an additional seven Atari 2600 video games. This expands its previous benchmark, reinforcing its capabilities in complex reinforcement learning environments. Wed, 11 Mar 2026 10:30:00 GMT Benchmark Update https://world-models.io/en/timeline/dreamerv3-atari-benchmark-update https://world-models.io/en/models/gamengen The first neural model to simulate a complex game (DOOM) in real-time at high quality, making the game engine itself a neural network. Wed, 11 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/gamengen https://world-models.io/en/timeline/meta-vjepa2-paper Meta FAIR introduces V-JEPA 2, extending the Joint Embedding Predictive Architecture to. The model learns representations by predicting abstract latent targets rather than pixels, achieving strong downstream performance on understanding benchmarks. Tue, 10 Mar 2026 09:45:00 GMT Paper Published https://world-models.io/en/timeline/meta-vjepa2-paper https://world-models.io/en/models/diamond DIffusion As a Model Of the eNvironment in Deep RL: uses diffusion models as world models for reinforcement learning agents. Tue, 10 Mar 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/diamond https://world-models.io/en/models/i-jepa Image Joint Embedding Predictive Architecture: learns visual representations by predicting abstract image regions without pixel reconstruction. Tue, 10 Mar 2026 00:00:00 GMT Self-Supervised World Model https://world-models.io/en/models/i-jepa https://world-models.io/en/research/model-based-rl What model-based reinforcement learning is, how world models enable imagination-based planning, and why Dreamer, MuZero, PlaNet, and TD-MPC2 matter. Tue, 10 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/model-based-rl https://world-models.io/en/compare/dreamer-v3-vs-planet DreamerV3 represents the evolution of PlaNet's core ideas. Both use the RSSM architecture, but DreamerV3 adds discrete representations, symlog predictions, and fixed hyperparameters to achieve state-of-the-art performance across diverse domains. Tue, 10 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-planet https://world-models.io/en/compare/i-jepa-vs-mae I-JEPA and MAE are both self-supervised image learning methods, but they follow opposite philosophies: I-JEPA predicts in abstract representation space, while MAE reconstructs masked pixels. Tue, 10 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/i-jepa-vs-mae https://world-models.io/en/timeline/stanford-survey-world-models-embodied This survey paper formalizes the problem setting and learning objectives of world models for embodied AI, which enable agents to perceive, act, and anticipate. It proposes a three-axis taxonomy based on functionality, temporal modeling, and spatial representation. Sun, 08 Mar 2026 12:00:00 GMT Paper Published https://world-models.io/en/timeline/stanford-survey-world-models-embodied https://world-models.io/en/models/unisim A universal simulator that learns to simulate real-world interactions from diverse data sources. Sun, 08 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/unisim https://world-models.io/en/models/copilot4d A world model for autonomous driving that predicts future LiDAR point clouds in 4D (3D space + time) using discrete diffusion. Sun, 08 Mar 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/copilot4d https://world-models.io/en/research/self-supervised-world-models How self-supervised world models learn environment dynamics without rewards, from JEPA and V-JEPA to predictive latent representations. Sun, 08 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/self-supervised-world-models https://world-models.io/en/compare/dreamer-v3-vs-muzero Both are landmark world model systems, but with fundamentally different architectures. DreamerV3 uses latent imagination with actor-critic learning, while MuZero uses abstract learned dynamics with Monte Carlo tree search. Sun, 08 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/dreamer-v3-vs-muzero https://world-models.io/en/timeline/editorial-comparator-methodology New features on world-models. io: the Model Comparator lets you compare any two world models side-by-side across architecture, benchmarks, and capabilities. The Performance Index methodology page details how models are scored and ranked. Fri, 06 Mar 2026 15:00:00 GMT Editorial https://world-models.io/en/timeline/editorial-comparator-methodology https://world-models.io/en/research/ai-simulation-systems How AI simulation systems and learned simulators reduce the reality gap and extend or replace hand-crafted engines for autonomous agents. Fri, 06 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/ai-simulation-systems https://world-models.io/en/models/td-mpc2 A scalable world model agent that combines TD-learning with model-predictive control across 104 diverse tasks. Thu, 05 Mar 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/td-mpc2 https://world-models.io/en/models/emu-video Meta's efficient video generation model using a factorized approach: first generate an image, then animate it into a video. Thu, 05 Mar 2026 00:00:00 GMT Generative World Model https://world-models.io/en/models/emu-video https://world-models.io/en/research/world-models-vs-llms The key differences between world models and LLMs across objective, architecture, planning, physical reasoning, and embodied AI use cases. Thu, 05 Mar 2026 00:00:00 GMT Research https://world-models.io/en/research/world-models-vs-llms https://world-models.io/en/compare/world-models-vs-llms World models and LLMs represent fundamentally different approaches to AI. World models learn causal dynamics of physical environments; LLMs learn statistical patterns over text. Both are essential for the future of AI. Thu, 05 Mar 2026 00:00:00 GMT Comparison https://world-models.io/en/compare/world-models-vs-llms https://world-models.io/en/timeline/world-models-2026-race-editorial In 2026, the artificial intelligence landscape saw a significant shift as world models began to eclipse large language models (LLMs) in prominence. DeepMind's Genie 3 launched as the first real-time, interactive, physically accurate 3D world model, capable of building environments from text or images. Mon, 02 Mar 2026 12:00:00 GMT Editorial https://world-models.io/en/timeline/world-models-2026-race-editorial https://world-models.io/en/timeline/leaderboard-march-2026-update The world-models. io Performance Index has been recalculated using the latest benchmark data. Key changes: NVIDIA Cosmos enters the top 5 for generative world models; DreamerV3 maintains its lead in model-based RL. Mon, 02 Mar 2026 10:00:00 GMT Leaderboard Update https://world-models.io/en/timeline/leaderboard-march-2026-update https://world-models.io/en/models/gaia-1 A generative world model for autonomous driving that predicts realistic driving scenarios from text, action, and video inputs. Sun, 01 Mar 2026 00:00:00 GMT Foundation World Model https://world-models.io/en/models/gaia-1 https://world-models.io/en/timeline/collection-autonomous-driving This resource introduces "World Models for Autonomous Driving," a new collection comprising 12 key models. It functions as a knowledge hub, tracking the release of world models, research papers, and benchmark updates across various AI domains. Sat, 28 Feb 2026 14:00:00 GMT Collection Update https://world-models.io/en/timeline/collection-autonomous-driving https://world-models.io/en/timeline/toyota-diffusebot Toyota Research Institute introduces DiffuseBot, diffusion-based world model that co-optimizes soft robot morphology and control. The model learns to simulate deformable body dynamics and generates novel robot designs optimized for specific tasks. Wed, 25 Feb 2026 10:00:00 GMT Paper Published https://world-models.io/en/timeline/toyota-diffusebot https://world-models.io/en/models/iris A world model agent that tokenizes observations into discrete tokens and models environment dynamics autoregressively, like a language model over game frames. Wed, 25 Feb 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/iris https://world-models.io/en/models/muzero Masters games without knowing the rules by learning a world model for planning via Monte Carlo tree search. Fri, 20 Feb 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/muzero https://world-models.io/en/models/dreamer-v2 The first model-based agent to achieve human-level performance on the Atari benchmark using discrete world model representations. Wed, 18 Feb 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/dreamer-v2 https://world-models.io/en/models/ha-world-model The seminal paper that popularized the concept of world models: learning to imagine environments and training policies entirely in dreams. Tue, 10 Feb 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/ha-world-model https://world-models.io/en/models/planet Deep Planning Network: learns environment dynamics in latent space for image-based control without a policy network. Thu, 05 Feb 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/planet https://world-models.io/en/models/rssm Recurrent State-Space Model: the foundational architecture behind PlaNet and the entire Dreamer family of world models. Thu, 05 Feb 2026 00:00:00 GMT Latent Dynamics https://world-models.io/en/models/rssm https://world-models.io/en/timeline/project-genie-public-launch Project Genie, an experimental research prototype, is now accessible to Google AI Ultra subscribers in the U. S. This prototype enables users to create, explore, and remix interactive worlds using text prompts and images. Thu, 29 Jan 2026 15:00:00 GMT Model Release https://world-models.io/en/timeline/project-genie-public-launch https://world-models.io/en/models/predictron An architecture that integrates learning and planning into a single differentiable network via abstract world models. Wed, 28 Jan 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/predictron https://world-models.io/en/timeline/world-labs-blog-announcing-the-world-api-1778612871923 World Labs introduces the World API, a public interface for generating explorable 3D worlds from various inputs like text, images, and video, powered by their Marble world model. This API allows for the programmatic creation of navigable spatial environments, aiming to make spatial intelligence universally accessible. Wed, 21 Jan 2026 12:00:00 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-announcing-the-world-api-1778612871923 https://world-models.io/en/models/imagination-augmented-agents An agent architecture that augments model-free policies with learned imagination rollouts from an environment model. Tue, 20 Jan 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/imagination-augmented-agents https://world-models.io/en/models/value-prediction-network A neural network that learns to plan by predicting future values and rewards through abstract state transitions, without reconstructing observations. Thu, 15 Jan 2026 00:00:00 GMT Model-Based RL https://world-models.io/en/models/value-prediction-network https://world-models.io/en/timeline/runway-research-introducing-runway-gwm-1-1773702004613 Runway introduces GWM-1, family of General World Models designed for real-time simulation and interaction with reality. Built on Gen-4. 5, GWM-1 operates autoregressively, generating frame-by-frame that is interactively controllable. Thu, 11 Dec 2025 16:22:35 GMT Model Release https://world-models.io/en/timeline/runway-research-introducing-runway-gwm-1-1773702004613 https://world-models.io/en/timeline/runway-research-gwm-1-a-state-of-the-art-general-world-model-built-to-interact-w Runway introduces GWM-1, a family of state-of-the-art Generative World Models designed to simulate and interact with reality in real-time. Built on Gen-4. 5, GWM-1 operates autoregressively, frame by frame, and allows interactive control via actions like camera pose and robot commands. Thu, 11 Dec 2025 16:22:35 GMT Model Release https://world-models.io/en/timeline/runway-research-gwm-1-a-state-of-the-art-general-world-model-built-to-interact-w https://world-models.io/en/timeline/world-labs-blog-marble-world-model-1773702003778 The Marble multimodal world model generates and simulates 3D worlds from various inputs including text, images, videos, and coarse 3D layouts. Users can interactively edit, expand, and combine these worlds, which can then be exported as Gaussian splats, meshes, or videos. Wed, 12 Nov 2025 10:00:00 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-marble-world-model-1773702003778 https://world-models.io/en/timeline/world-labs-blog-marble-world-model-1778572725540 Marble is a new multimodal world model for creating 3D environments from various inputs. It reconstructs, generates, and simulates 3D worlds, allowing for human and agent interaction. Wed, 12 Nov 2025 10:00:00 GMT Lab Update https://world-models.io/en/timeline/world-labs-blog-marble-world-model-1778572725540 https://world-models.io/en/timeline/genie-3-announcement Google DeepMind unveiled Genie 3, a world model capable of generating diverse, interactive 3D environments from text prompts. This AI system can create dynamic worlds navigable in real time at 24 frames per second, maintaining consistency for several minutes at 720p resolution. Tue, 05 Aug 2025 12:00:00 GMT Model Release https://world-models.io/en/timeline/genie-3-announcement https://world-models.io/en/timeline/vjepa2-paper-arxiv V-JEPA 2 is a self-supervised video model pre-trained on over 1 million hours of internet video. It achieves strong performance in motion understanding (77. 3% on Something-Something v2) and state-of-the-art human action anticipation (39. 7% recall-at-5 on Epic-Kitchens-100). Wed, 11 Jun 2025 10:00:00 GMT Paper Published https://world-models.io/en/timeline/vjepa2-paper-arxiv https://world-models.io/en/timeline/kaist-world-model-benchmark The HistogramTools R package enhances histogram functionality for large-scale data analysis. It provides advanced methods for manipulating, merging, and analyzing histograms, including serialization via Protocol Buffers for distributed computing. Wed, 02 Apr 2025 10:00:00 GMT Benchmark Update https://world-models.io/en/timeline/kaist-world-model-benchmark https://world-models.io/en/timeline/cosmos-tokenizer-paper NVIDIA introduces Cosmos Tokenizer, a novel suite of visual tokenizers for images and videos that significantly advances generative AI. This system offers both continuous and discrete tokenization, achieving superior compression rates and reconstruction quality up to 12 times faster than existing methods. Mon, 10 Mar 2025 11:00:00 GMT Paper Published https://world-models.io/en/timeline/cosmos-tokenizer-paper https://world-models.io/en/timeline/genie2-public-demo-interactive Google DeepMind introduces Genie 2, a large-scale foundation world model capable of generating diverse, interactive 3D environments from a single prompt image. Designed for training and evaluating embodied AI agents, Genie 2 allows human or AI agents to control characters using keyboard and mouse inputs. Mon, 17 Feb 2025 15:00:00 GMT Model Release https://world-models.io/en/timeline/genie2-public-demo-interactive https://world-models.io/en/timeline/lwm-large-world-model-paper This paper introduces GhostWriter, an AI-enhanced design probe aiming to improve personalization and agency in human-AI collaborative writing. GhostWriter implicitly learns user writing style using LLMs and offers explicit teaching moments for refinement. Mon, 04 Nov 2024 10:00:00 GMT Paper Published https://world-models.io/en/timeline/lwm-large-world-model-paper https://world-models.io/en/timeline/diamond-diffusion-world-model DIAMOND introduces a reinforcement learning agent trained within a diffusion-based world model, challenging traditional discrete latent variable models. This approach demonstrates that superior visual detail in dynamics modeling leads to improved agent performance. DIAMOND achieved a mean human-normalized score of 1. Tue, 15 Oct 2024 10:00:00 GMT Model Release https://world-models.io/en/timeline/diamond-diffusion-world-model https://world-models.io/en/timeline/rt2-robotic-transformer-release Google DeepMind introduces RT-2, a novel Vision-Language-Action (VLA) model that enhances robotic control by integrating knowledge from both web-scale datasets and robotics data. Building upon the RT-1 model, RT-2 translates generalized instructions for robotic control while retaining web-scale capabilities. Sun, 28 Jul 2024 15:00:00 GMT Model Release https://world-models.io/en/timeline/rt2-robotic-transformer-release https://world-models.io/en/timeline/lecun-ami-world-models-update Yann LeCun's paper, A Path Towards Autonomous Machine Intelligence, outlines an architectural and training paradigm for autonomous intelligent agents. Wed, 12 Jun 2024 15:00:00 GMT Paper Published https://world-models.io/en/timeline/lecun-ami-world-models-update https://world-models.io/en/timeline/google-veo-video-generation Google DeepMind introduces Veo, state-of-the-art generation model capable of producing 1080p videos beyond 60 seconds. Veo demonstrates strong understanding of physics, human movement, and cinematic styles, positioning it as direct competitor to OpenAI Sora in the emerging world-model-as- -generator paradigm. Tue, 14 May 2024 17:00:00 GMT Model Release https://world-models.io/en/timeline/google-veo-video-generation https://world-models.io/en/timeline/runway-research-introducing-general-world-models-1773702004540 Runway Research introduces general world models (GWMs) as the next major advancement in AI. GWMs are AI systems that build internal representations of the visual world and simulate future events across diverse real-world situations, unlike current limited world models. Mon, 11 Dec 2023 10:10:10 GMT Model Release https://world-models.io/en/timeline/runway-research-introducing-general-world-models-1773702004540 https://world-models.io/en/timeline/runway-research-general-world-models-our-long-term-research-effort-to-build-ai-s This article introduces the concept of General World Models (GWMs) as the next frontier in AI research. GWMs are AI systems designed to build internal representations of the visual world and simulate future events across a wide range of real-world scenarios. Mon, 11 Dec 2023 10:10:10 GMT Model Release https://world-models.io/en/timeline/runway-research-general-world-models-our-long-term-research-effort-to-build-ai-s https://world-models.io/en/timeline/stability-stable-video-diffusion Stability AI open-sourced Stable Diffusion (SVD), latent diffusion model capable of generating short clips from single images. SVD established new baseline for open- generation and inspired numerous downstream world model research projects. Tue, 21 Nov 2023 14:00:00 GMT Model Release https://world-models.io/en/timeline/stability-stable-video-diffusion https://world-models.io/en/timeline/wayve-gaia-1-release Wayve introduces GAIA-1, a 9-billion parameter generative world model for autonomous driving. This model can generate realistic driving videos from various inputs, including video, text, and actions, offering fine-grained control over vehicle behavior and scene features. Thu, 29 Jun 2023 11:00:00 GMT Model Release https://world-models.io/en/timeline/wayve-gaia-1-release