Cosmos3-Nano (AI model): news timeline & CVEs

Into the Omniverse: How Developers Turn Ideas Into Simulations With Frontier AI Agents

NVIDIA showed developers using frontier AI agents with Omniverse libraries to build robotics, driving, and digital-twin simulations.

An NVIDIA developer blog shows engineers directing frontier AI agents, including GPT-6 Astra and Claude Fable 5, to build simulations with Omniverse libraries for physics, rendering, and sensors. Examples include a warehouse humanoid simulator, a San Francisco Market Street autonomous-driving testbed called Zero to Alpamayo, and digital-twin refinement by comparing simulated and recorded camera and LiDAR outputs. Other demos cover a Unitree G1 that cleared a hurdle in 64 of 100 physics trials, robotic suspension disassembly in PTC Onshape and Isaac Sim, and a browser scene of the International Space Station assembled from NASA assets.

NVIDIA Blog21:06 UTC · 2h agoAI tools & infra

NVIDIA Researchers Introduce Physis-Lang: Self-Evolving Physical Language That Lifts Cosmos 3 Past Veo 3.1 on Physics Benchmarks

NVIDIA's Physis-Lang physics captions lift Cosmos 3 past Veo 3.1 on several physics video benchmarks.

NVIDIA, MIT, and University of Oxford researchers introduce Physis-Lang, a self-evolving natural-language representation of physics used for video-data curation, training captions, and inference prompts. On the September 29, 2026 Physics-IQ Verified snapshot, Cosmos3-Super with Physis-Lang scored 48.2 and Cosmos3-Nano scored 43.3; Cosmos3-Nano also beat Veo 3.1 on PhyGenBench (71.04 vs 65.63) and Physics-IQ Verified (43.41 vs 34.99) after LoRA fine-tuning. Distilling the captioning pipeline into Qwen3-VL-4B-Instruct models preserved most of a 7.05-point Wan2.1-14B gain at roughly $120 instead of about $24,120 in API cost. The comparison lists the work as paper-only as of September 30, 2026.

MarkTechPostSep 30, 07:32 UTC · 8d agoAI research

Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model

Physis-Lang improves physical video generation, and Cosmos3-Nano variants surpass proprietary Veo 3.1.

Physis-Lang represents physical processes in language covering entities, causes, interactions, principles, evolution, and effects, then optimizes that language across curation, training, and generation. PhysCapBench decomposes processes into atomic assertions and scores caption recall and precision, while an agentic loop fixes assertion errors and retrieves videos for missing processes. On four physical-video benchmarks, Wan and Cosmos backbones improve, and Cosmos3-Nano variants enhanced by Physis-Lang surpass proprietary Veo 3.1.

Hugging Face daily papersSep 29, 20:00 UTC · 9d agoAI research1

Cosmos3-Nano

Timeline3 articles

Appears with

Into the Omniverse: How Developers Turn Ideas Into Simulations With Frontier AI Agents

NVIDIA Researchers Introduce Physis-Lang: Self-Evolving Physical Language That Lifts Cosmos 3 Past Veo 3.1 on Physics Benchmarks

Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model