{"id":"4ef34eaa-6694-4b8b-8492-9d14bbbd0703","slug":"nvidia-cosmos","name":"NVIDIA Cosmos","description":"NVIDIA's open world foundation model for physical AI. Cosmos 3 (launched May 2026 at GTC Taipei) uses a mixture-of-transformers architecture combining vision reasoning, world generation, and action prediction. The world's first fully open omnimodel — natively understands and generates text, image, video, ambient sound, and actions. Variants: Cosmos 3 Super (highest accuracy), Cosmos 3 Nano (fast), Cosmos 3 Edge (4B params, on-device). #1 on Artificial Analysis, Physics-IQ, PAI-Bench, R-Bench, RoboArena, VANTAGE-Bench.","brainType":"world-model","isOpen":true,"maturityStage":"commercial","architecture":"Transformer-based world model with text-to-world and video-to-world prediction. Cosmos-3 Edge runs on single H200 GPU.","reviewStatus":"reviewed","sources":[{"url":"https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-3-the-open-frontier-foundation-model-for-physical-ai","date":"2026-05-31","title":"NVIDIA Launches Cosmos 3, the Open Frontier Foundation Model for Physical AI","sourceName":"NVIDIA Newsroom"},{"url":"https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development","date":"2025-01-06","title":"NVIDIA Launches Cosmos World Foundation Model Platform","sourceName":"NVIDIA Newsroom"},{"url":"https://www.nvidia.com/en-us/ai/cosmos/","title":"Physical AI with World Foundation Models","sourceName":"NVIDIA"},{"url":"https://docs.nvidia.com/cosmos/latest/cosmos3/index.html","title":"Cosmos 3","sourceName":"NVIDIA Developer Documentation"},{"url":"https://research.nvidia.com/labs/cosmos-lab/cosmos3/technical-report.pdf","title":"Cosmos 3: Omnimodal World Models for Physical AI","sourceName":"NVIDIA Research"},{"url":"https://docs.nvidia.com/cosmos/latest/introduction.html","title":"Introduction","sourceName":"NVIDIA"}],"keyFacts":[{"label":"Cosmos 3 launch (primary NVIDIA)","value":"NVIDIA's May 31 2026 newsroom release, datelined GTC Taipei, says the company launched Cosmos 3 as an open world foundation model for physical AI on a mixture-of-transformers architecture that combines vision reasoning, world generation, and action prediction. NVIDIA calls Cosmos 3 the world's first fully open omnimodel that can natively understand and generate text, images, video, ambient sound, and actions. Treat the 'world's first' and leaderboard claims as NVIDIA-stated (NVIDIA Newsroom)"},{"label":"Cosmos 3 lineup and availability (primary NVIDIA)","value":"The same release says Cosmos 3 Super and Cosmos 3 Nano are available now, with Cosmos 3 Edge coming soon for real-time inference. Developers can try Cosmos 3 on build.nvidia.com, download open models from Hugging Face, and deploy them as NVIDIA NIM microservices. Super is described for highest physics accuracy; Nano for faster video and action reasoning; Edge for edge inference. Frame availability as NVIDIA-stated at launch (NVIDIA Newsroom)"},{"label":"Cosmos Coalition (primary NVIDIA)","value":"NVIDIA also announced the Cosmos Coalition, naming founding members Agile Robots, Black Forest Labs, Generalist, LTX, Runway, and Skild AI. The release lists additional builders on the platform including Doosan Robotics, LG Electronics, Samsung Electronics, and Li Auto. Treat membership as NVIDIA-stated (NVIDIA Newsroom)"},{"label":"World foundation model platform","value":"NVIDIA describes Cosmos as a platform of generative world foundation models for developing physical AI systems in robotics and autonomous machines (NVIDIA Cosmos launch)."},{"label":"Physical AI development","value":"NVIDIA's Cosmos product page says the platform provides world models and tools for physical AI development across robotics and autonomous systems (NVIDIA Cosmos)."},{"label":"Cosmos 3 modalities","value":"NVIDIA's Cosmos 3 documentation says the family jointly processes and generates language, images, video, audio, and action sequences using a unified Mixture-of-Transformers architecture for Physical AI (Cosmos 3 documentation)."},{"label":"Cosmos 3 open research release","value":"The Cosmos 3 technical report says NVIDIA is releasing code, checkpoints, curated synthetic datasets, and an evaluation benchmark under the OpenMDW-1.1 license; availability and license are NVIDIA-stated (Cosmos 3 technical report)."},{"label":"Physical AI foundation models","value":"NVIDIA describes Cosmos as a developer-first platform with world foundation models, video tokenizers, and data-processing tools for physical AI systems such as robots and autonomous vehicles (Cosmos introduction)."},{"label":"World-model platform","value":"NVIDIA says Cosmos supports physical AI development through generative world models and tools for predicting and reasoning about future physical states (NVIDIA Cosmos launch)."}],"aliases":[],"collisionRisk":"low","reviewNote":null,"builtOnBrainId":null,"createdAt":"2026-08-09T14:35:22.669Z","updatedAt":"2026-10-09T22:00:40.123Z","jsonLd":{"@context":"https://schema.org","@type":"SoftwareApplication","@id":"https://registry.deploy.report/brains/nvidia-cosmos","url":"https://registry.deploy.report/brains/nvidia-cosmos","name":"NVIDIA Cosmos","description":"NVIDIA's open world foundation model for physical AI. Cosmos 3 (launched May 2026 at GTC Taipei) uses a mixture-of-transformers architecture combining vision reasoning, world generation, and action prediction. The world's first fully open omnimodel — natively understands and generates text, image, video, ambient sound, and actions. Variants: Cosmos 3 Super (highest accuracy), Cosmos 3 Nano (fast), Cosmos 3 Edge (4B params, on-device). #1 on Artificial Analysis, Physics-IQ, PAI-Bench, R-Bench, RoboArena, VANTAGE-Bench.","identifier":"4ef34eaa-6694-4b8b-8492-9d14bbbd0703","applicationCategory":"world-model","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"verified","maturity_stage":"commercial","lifecycle_state":null,"architectural_position":{"cohort":null,"sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":6,"primary_source_types":[]}}}