Nvidia Expands Cosmos 3 With Open-Source Robotics Tools and Humanoid Robot Reference Design
Nvidia expanded its Cosmos 3 rollout with a large set of open-source tools for agentic and physical AI development and what it described as its first open humanoid robot reference design for robotics research. The broader package builds on the earlier introduction of Cosmos 3, a fully open omnimodel for physical AI with native vision reasoning, world generation and action generation, in Super and Nano variants at 32B and 8B parameters.
Nvidia said Cosmos 3 uses what it calls a MoT architecture, pairing an autoregressive reasoner tower with a diffusion-based generator tower and combining functions that had previously been handled separately. The company said the model was trained on billions of samples across modalities and can simulate physical environments, predict future world states and help train robots with less data and lower training costs. Nvidia said model weights and post-training recipes are available on Hugging Face, and executives said it plans partnerships with humanoid robot makers in the U.S., Europe, South Korea and China, including Unitree.
From the sources (24 posts)
@nvidiaaiIntroducing Cosmos 3: Our latest frontier model for Physical AI Cosmos 3 is the world’s first fully open omnimodel with native vision reasoning, world and action generation. Today we’re releasing Super (32B) and Nano (8B) variants. https:
@nvidiaaiCosmos 3 ties everything together. Previous releases separated world generation, physical understanding, and controlled scene generation. Cosmos 3’s MoT architecture unifies these capabilities by pairing an autoregressive reasoner tower w
@nvidiaaiDelivering leading results on physical AI benchmarks among open models, it ranks first across @ArtificialAnlys, Physics-IQ, PAI-Bench and R-Bench for world generation accuracy, RoboLab and RoboArena for action policy and the VANTAGE-Bench a
@nvidiaai@ArtificialAnlys In addition to understanding and reasoning across modalities, Cosmos 3 excels at simulating physical environments, predicting future world states, and helping train robots to perform specific tasks. It can do subsecond vi
@theturingpostBreaking news: Cosmos 3 is here. They are attempting to do something completely new 🤯 Why is Physical AI much harder than building a chatbot? Understanding the world is not enough, robots need to predict it and act inside it. That's the
@nvidiaai@ArtificialAnlys Image-to-video generation is just as impressive. Input image: "Generate a 16:9 image from a dashcam view of a formula 1 racing event" Video prompt: "A high-speed racing event where a car navigates multiple winding turns"
@financialjuiceNvidia CEO: introduces new AI model Nemotron 3 Ultra at Taiwan Computex
@firstsquawkJensen Huang has launched Nemotron 3 Ultra, a new AI model, during a presentation at Computex Taiwan.
@firstsquawkNVIDIA introduces a new AI agent toolkit built around Nemoclaw, Nemotron, OpenShell, and CUDA-X.
@firstsquawkJensen Huang announces that NVIDIA is launching a new model generation designed for robotics-based artificial intelligence.
@nvidiaai@ArtificialAnlys Trained on billions of samples across modalities, the model provides developers with a powerful pretrained foundation for building physical AI systems with less data and lower training costs. Read more in our technical blo
@firstsquawkNVIDIA has launched a large set of open-source tools and capabilities for agentic and physical AI development.
@nvidiaai@ArtificialAnlys As always, Cosmos 3 is fully open. This includes model weights and post-training recipes. Available now on @huggingface
@financialjuiceNvidia unveils extensive open source toolkit for physical AI agents
@cnbcNvidia picks Unitree for humanoid robot platform as Chinese startup eyes IPO
@_akhaliqRT @NVIDIAAI: @ArtificialAnlys As always, Cosmos 3 is fully open. This includes model weights and post-training recipes. Available now on…
@firstsquawkNVIDIA plans partnerships with humanoid robot makers in multiple regions, including the U.S., Europe, South Korea, and China’s Unitree, executives said.
@stockmktnewzNvidia $NVDA just posted this: “The first open humanoid robot reference design built for robotics research.”
@basetenWant to teach a robot to open a door without shattering the glass, crushing the handle, or clipping through reality? If so, NVIDIA Cosmos 3 might be the most important model release of the year for you. Cosmos 3 is designed to respect cons
@vllm_project🚀 Excited to partner with @NVIDIAAI on day-0 support for Cosmos 3 on vLLM-Omni! A unified Mixture-of-Transformers fusing an AR reasoner + diffusion generator across text, image, video, audio & robot action - all behind a single OpenAI-comp
@mervenoyannNVIDIA just dropped Cosmos 3 at GTC 🔥 closest thing to AGI as world model > it can reason, understand AND generate videos, images, actions, text > sota, comes in 16B, 65B, with datasets > diffusers support 🧨 > open license 🤗
@cointelegraph🔥 TODAY: Nvidia unveils its first open humanoid robot reference design for robotics research, combining a full-stack platform from data capture to model deployment.
@victormustarIn case you missed this: NVIDIA shipped a text-to-image open weights model that looks seriously competitive 👀 (as part of its Cosmos 3 release)
@theturingpostRT @TheTuringPost: Breaking news: Cosmos 3 is here. They are attempting to do something completely new 🤯 Why is Physical AI much harder th…