{"id":"cosmos3-h-embodied-nano","slug":"cosmos3-h-embodied-nano","name":"Cosmos3-H-Embodied-Nano","orgId":"nvidia","family":"cosmos3-h-embodied-nano","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://huggingface.co/nvidia/Cosmos3-H-Embodied-Nano","hfRepo":null,"modalities":["robotics","video"],"aliases":["Cosmos3-H-Embodied-Nano","cosmos3-h-embodied-nano","cosmos3 h embodied nano","nvidia/cosmos3-h-embodied-nano","cosmos3 h-embodied nano"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-20","announced":"2026-10-20","blurb":"NVIDIA's 16B surgical-robotics model for dVRK-style and other surgical robots. One checkpoint predicts robot actions and endoscopic video, serving as world model (forward dynamics), policy and inverse dynamics model. Open weights under the NVIDIA Open Model License; the card says it is ready for commercial use.","api":"none","openWeights":"yes","params":{"text":"16B (Cosmos3-Nano size; about 7B updated in post-training)","source":"NVIDIA's Hugging Face model card"},"facts":[{"label":"Type","value":"Mixture-of-Transformers world foundation model for surgical robotics: forward dynamics, inverse dynamics and policy in one model"},{"label":"Base model","value":"Cosmos3-Nano (Qwen3-VL-8B reasoning tower plus a diffusion generation expert), post-trained for 50,000 iterations"},{"label":"Training data","value":"Open-H-Embodiment surgical robotics data, 15 dataset families (11 public on Hugging Face)"},{"label":"Clips","value":"2.4 s: 25 frames and 24 actions at 10 Hz, 256p resolution tier"},{"label":"License","value":"NVIDIA Open Model License; ready for commercial or non-commercial use (model card)"},{"label":"Predecessor","value":"Cosmos-H-Surgical-Simulator (forward dynamics only, 2B)"},{"label":"Hugging Face repo created","value":"Oct 19, 2026 (UTC); the card lists Oct 20, 2026 as the release date"}],"claims":["Lab claim: on held-out clips, inverse-dynamics action error is 8.6× below a mean-trajectory baseline, and forward-dynamics video beats Cosmos-H-Surgical-Simulator on dVRK per-pixel metrics."],"sources":[{"kind":"repo","title":"Cosmos3-H-Embodied-Nano on Hugging Face","url":"https://huggingface.co/nvidia/Cosmos3-H-Embodied-Nano","date":"2026-10-20"},{"kind":"repo","title":"Cosmos3-H-Embodied-Nano on GitHub (NVIDIA-Medtech)","url":"https://github.com/NVIDIA-Medtech/Cosmos3-H-Embodied-Nano","date":"2026-10-20"},{"kind":"press","title":"HuggingPapers post on X","url":"https://x.com/HuggingPapers/status/2112599188899045456","date":"2026-10-20"}],"firstBenchmark":null},"organization":{"id":"nvidia","name":"NVIDIA","slug":"nvidia","country":null,"url":null,"color":"#76B900","mark":"N"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 20, 2026; no independent results yet.","lead":"As of Oct 22, 2026, Cosmos3-H-Embodied-Nano is pending benchmarks on isitsota: announced Oct 20, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}