I joined Daniel and Chris on Practical AI to discuss open models, NVIDIA Cosmos, and how world models can help AI understand and simulate the physical world. We explore why openness matters and what it takes to build increasingly capable physical AI systems. Watch: piped.video/watch?v=W1DSS-Sz… Listen: practicalai.show/374

AI is moving beyond the cloud and into robots, vehicles, and other physical systems. @liu_mingyu, VP of Cosmos Lab at @NVIDIAAI, joins us to discuss open models, world models, simulation, and what’s next for physical AI. 🎧 Listen/Watch: [link below]

In Cosmos, we build world foundation models for physical AI. The term “world foundation model,” coined by @JensenHuang, brings together two ideas: world models and foundation models. We build models to solve tasks. Because we live in the physical world, many tasks require knowledge of the world—its objects, environments, and physical behavior. Here, we use “world model” in a broad sense to describe a model that captures the knowledge of the world needed to solve a task, rather than adopting a specific definition from the robotics literature. Different tasks require different kinds of knowledge, so we build different world models, each using data suited to its purpose. Different tasks call for different world models, but those models all describe aspects of the same physical world. This shared basis makes it possible to bring together the diverse data used to train specialized world models and build a single model that learns across them. A world foundation model is, in this sense, a foundation model of world models: it learns shared knowledge of the physical world that can be adapted to many tasks and environments. This matters because we are entering the era of physical AI, where robots and other autonomous systems will perform useful tasks for us in the real world. These systems must operate in complex, varied, and constantly changing environments. World foundation models can help them anticipate the consequences of actions, evaluate possible futures, and make better decisions. That is the goal behind our work in Cosmos.

Thanks, Nate B. Jones, for a thoughtful discussion on Cosmos 3, simulation and verifiable rewards for physical AI. Moving more development into simulation can help teams test faster, learn sooner and bring better systems into the real world. When Will AI Make Me Scrambled Eggs? I Went To NVIDIA To Find Out. piped.video/ry9J1i3krIY?is=n5Ln… via @YouTube

Looking forward to joining the Runway AI Summit on Sept. 30 in San Francisco. I would like to share our perspective on how the world of Physical AI might evolve and how we would like to support the ecosystem. If you’re working on physical AI, join us. Register: summit.runwayml.com/

The Runway AI Summit is one week from today. Announcing our sessions: Grounding Intelligence in the Physical World Ming-Yu Liu of Nvidia; Jon Barron from Google DeepMind; Alex Toshev from Wayve

NVIDIA Cosmos Lab is looking for PHD interns to help build Open Physical AI foundation models. The job is in Santa Clara, CA. Below is the application link. nvidia.wd5.myworkdayjobs.com…

I had a great conversation with @MLStreetTalk about world models, Physical AI, and Cosmos 3—and how we can help the ecosystem build more generalizable Physical AI with better data, better environments, and better starting points. Really enjoyed the discussion and the chance to share where we see world models heading next.

How Physical AI Learns Across Language, Video and Action — Ming-Yu Liu piped.video/L6tLBApQN-g?is=funF… via @YouTube

Today we're excited to announce Mercury 2.5 It’s the most capable diffusion LLM on the market. It is a 40% jump in intelligence over Mercury 2 and runs at over 1,100 tokens/sec on widely available @NVIDIAAI GPUs. inceptionlabs.ai/blog/introd…

What if a dexterous robot could learn a 20+ step chemistry experiment without a single on-robot training demo? Meet TwinDEX: a pair of co-designed, three-finger, nine-DoF dexterous manipulation interface: one wearable for data collection, one for robot deployment. The twinned design shares identical kinematics, contact surfaces, visual appearance, and sensors across collection and deployment — keeping observations and actions aligned end to end. Trained from scratch on only a few hundred wearable demonstrations - with zero on-robot training or intervention data - TwinDEX completed a standardized chemistry experiment involving tool switches, fine force control, and bimanual coordination. Robot-free data showed comparable learning efficiency on the multi-task evaluation, TwinDEX delivered 5.3 times effective throughput than on-robot teleoperation. TwinDEX demonstrates that high-quality robot-free data can fully substitute for on-robot teleoperation data on challenging dexterous tasks — removing the dependency on real-robot hardware that has been the central bottleneck to scaling dexterous manipulation data. This was the proof-it phase. Now comes scale: what emerges at tens of thousands, or millions, of episodes? Watch the demo and read the technical blog: x2robot.com/en/pages/twindex #TwinDEX #Robotics #EmbodiedAI #DexterousManipulation

Looking forward to Fellows Forum 2026. On Sept. 24 at 9:45 a.m. PT, I’ll give a keynote on foundation models for robotics and the physical world. If you’re interested in physical AI, please join us in Menlo Park or via livestream.

Sept 23-24 in Menlo Park is our Fellows Forum: a private gathering of 500 top founders, researchers, enterprise leaders, and investors. Two days of technical depth, keynotes, demos, and meetings. We're partnering with @NebiusAI and @NVIDIA to bring this year's event to another level.

"Data stops being something you collect. It becomes something you compute." At #SIGGRAPH2026, @liu_mingyu, VP of NVIDIA's Cosmos Lab, laid out why world models are the data engine of physical AI. Plus, he announced Cosmos-Dreams, a neural closed-loop simulator, and showed off a full demo of it in action. You can watch the full keynote here: vist.ly/5cwd9

10 million downloads! 🎉 An incredible milestone for NVIDIA Cosmos—and a testament to the team’s vision, hard work, and commitment to advancing physical AI. Proud of what we’ve built together, and even more excited for what developers and researchers will create next. 💚

10 million downloads and counting for NVIDIA Cosmos models on @huggingface. 🤗 This milestone belongs to the developers and researchers using open world foundation models to build robots, autonomous vehicles and other physical AI systems. Thank you for downloading, experimenting, and building with us. 💚

Dreamers deserve tools they can actually build with. An API gives you answers. Open weights give you a foundation — something you can fine-tune, tear apart, learn from, and make yours. That's how one model becomes ten thousand things its creators never imagined: a medical assistant in a language the original team didn't speak, a robot policy trained in a garage, a research idea tested overnight instead of never. Closed models scale usage. Open models scale creation.

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…

Robots need world models that can think and act at robot speed. ⚡ Cosmos 3 Edge on Jetson brings multimodal reasoning and action on-device, small enough for the smallest boards and fast enough for real robot control loops. It delivers higher speed and accuracy than similarly sized open models on the same Jetson hardware, making world models practical for robots, not just cloud demos.