NVIDIA Cosmos is a platform of world foundation models and development tools for physical AI, addressing challenges developers face in adapting models to their sensors, tasks, and data and in getting those models into applications reliably. The Cosmos Framework provides agent skills and recipes for world model training, multimodal data processing, optimization and inference, while Cosmos NIM microservices supports GPU-accelerated serving through application APIs. The framework is an open and fully modular framework for world model training, model optimization, and inference, and is expanding beyond Cosmos 3 to support NVIDIA Alpamayo, GR00T, Qwen, pi0.5, and other compatible Hugging Face models. Training supports text, images, video, audio, and actions, with LoRA for parameter-efficient adaptation or full supervised fine-tuning (SFT) to update all model parameters, and data loading through JSONL, WebDataset, LeRobot, or custom adapters. The platform covers online and offline RL paths, including DPO, GRPO, PPO, OPD, and DiffusionNFT. Models can be distilled to reduce model size or generation steps and quantized to reduce memory use, and run with Hugging Face Diffusers, PyTorch, SGLang, vLLM-Omni, vLLM, TRT-LLM and TRT-Edge-LLM or NVIDIA NIM.