NVIDIA releases four open Cosmos 3 Nano checkpoints for interactive world simulation
New open-weight models enable multi-sensor driving simulation, streaming video generation, and robotics interaction under permissive OpenMDW 1.1 license.
1 source · cross-referenced
- Four Cosmos 3 Nano open-weight checkpoints released for interactive world simulation and multi-sensor driving scenes.
- Models support synchronized RGB–LiDAR generation, streaming observations, simulator-to-video rendering, and bimanual robotics simulation.
- All models are post-trained variants of a 16B-parameter base and licensed under OpenMDW 1.1 for commercial and non-commercial use.
- Checkpoints and technical report are available on Hugging Face; demos and inference skills provided.
NVIDIA published four new open-weight checkpoints under the Cosmos 3 Nano family on Hugging Face, targeting interactive world simulation and multi-sensor driving scenes. The releases include Cosmos3-Nano-Transfer-Auto, Cosmos3-Nano-Sim-Auto, Cosmos3-Nano-Sim-Transfer, and Cosmos3-Nano-Sim-Bimanual, each designed for complementary simulation workflows.
Cosmos3-Nano-Transfer-Auto generates synchronized multi-view driving data across up to 11 camera views, including fisheyes, and supports joint RGB–LiDAR generation. It accepts World Scenario Map controls for scene structure and traffic motion, text prompts for appearance, and HD-map range controls for LiDAR, enabling developers to vary weather, lighting, and time-of-day conditions for training and evaluation coverage.
Cosmos3-Nano-Sim-Auto is a causally distilled streaming variant of Transfer-Auto that produces successive chunks of synchronized camera video and LiDAR observations in a few denoising steps. It uses generated history and updated driving controls to support fast, long interactive rollouts, requiring a simulation runtime and WSM renderer to update controls from driving inputs.
Cosmos3-Nano-Sim-Transfer renders photorealistic video from simulator spatial controls and a text prompt, producing outputs such as depth, segmentation, edges, and blur while delegating physics and ground-truth state to the simulator. This checkpoint connects physics simulators to generated RGB observations for policy consumption.
Cosmos3-Nano-Sim-Bimanual simulates how an environment evolves in response to a bimanual robot’s actions, supporting interactive two-arm simulation for YAM and AgiBot embodiments. It accepts mixed-action sequences with 64 normalized control values and requires aligned timing between actions and observations.
All four models are post-trained examples of Cosmos 3 Nano and share a 16B trainable parameter base. They are released under OpenMDW 1.1, permitting commercial and non-commercial use. The checkpoints, technical report, demos, and inference skills are available on Hugging Face, with notebooks and inference endpoints provided for immediate experimentation.
- Nov 11, 2026 · Hugging Face
Hugging Face proposes token pooling-aware training to improve compressed multi-vector retrieval
Trust79 - Nov 11, 2026 · TechCrunch — AI
Fractile raises $720M Series C to develop AI chips and compiler stack
Trust79 - Nov 11, 2026 · The Verge — AI
Google’s Tabletop AI experiment lets users build custom collaborative apps via waitlist
Trust79