Written by Jakub Rusinowski · Last updated July 21, 2026
PREVIEW (2026). NVIDIA's open frontier World Foundation Model (WFM) for Physical AI — the first fully open omnimodel that natively understands and generates text, images, video, ambient audio and robot/vehicle actions in a single system, with leading physics accuracy. Cosmos 3 is a two-tower Mixture-of-Transformers (MoT): a reasoner tower (an autoregressive vision-language "brain" that interprets images/video/text and reasons about motion, object interactions and physical cause-and-effect) plus a diffusion-based generator tower that predicts physics-aware future video and action sequences conditioned on that reasoning — so it "thinks before it acts." The two towers keep separate norm layers and MLPs but share multimodal attention. Trained on ~20 trillion tokens of multimodal data (nearly a billion images, 400M+ real and synthetic videos, plus audio and human/robot action data), it compresses physical-AI training and evaluation from months to days by generating synthetic worlds and predicting actions for robots and autonomous vehicles. Ships in three scales — Super (64B), Nano (16B) and Edge (4B) — under the permissive OpenMDW-1.1 license, on Hugging Face (nvidia/) and NVIDIA NGC. NVIDIA reports open-source state-of-the-art results on R-Bench, PAI-Bench, Physics-IQ and RoboLab; treat launch numbers as vendor-reported pending independent replication. This is a robotics / autonomous-vehicle / world-simulation model, not a chat LLM — verify specs on the model cards before relying on them.
| Cosmos 3 Super | Min 39 GB VRAM · BF16 · Omnimodal I/O (video/action) ctx · |
| Cosmos 3 Nano | Min 10 GB VRAM · BF16 · Omnimodal I/O (video/action) ctx · |
| Cosmos 3 Edge | Min 3 GB VRAM · BF16 · Omnimodal I/O (video/action) ctx · |
The cheapest GPU that runs Cosmos 3 locally (min 3 GB VRAM) is the Intel Arc B570 (10 GB).
Install Ollama then run: ollama run
Minimum VRAM: 3 GB. For best results use Q4_K_M quantization.
Cosmos 3 needs about 3 GB VRAM at BF16 quantization for its smallest variant. Variants: Cosmos 3 Super (39 GB, BF16); Cosmos 3 Nano (10 GB, BF16); Cosmos 3 Edge (3 GB, BF16). On Apple Silicon, unified memory counts toward this requirement.
Yes — Cosmos 3 runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at BF16.
Q4_K_M is the best balance of quality and VRAM for Cosmos 3 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run . This downloads Cosmos 3 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.