On 6 August 2026, NVIDIA announced Cosmos 3 Super, Cosmos 3 Nano and Cosmos 3 Edge. The family offered 64B Super, 16B Nano and 4B Edge checkpoints spanning cloud world modelling to on-device robot reasoning.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 6 August 2026 |
| Availability or weight release | 6 August 2026 as an open model collection, code and technical report |
| Release type | open-weight physical-AI world foundation-model family |
| Access | downloadable under OpenMDW 1.1 with Hugging Face, GitHub and NVIDIA tooling |
| Architecture | mixture-of-transformers omni-model family for vision reasoning, world generation and world-action prediction |
| Maximum stated context | physical-scene and video context varies by checkpoint and task rather than one launch token headline |
What changed
Cosmos 3 consolidated scene understanding, synthetic-data generation, future-state simulation and action modelling into one open family. It also created a practical scale ladder from frontier generation to edge deployment, backed by an official model collection and technical report.
The practical comparison is therefore not simply whether Cosmos 3 Super, Cosmos 3 Nano and Cosmos 3 Edge has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Open-weight text-to-image and image-to-video | ranked first by Artificial Analysis at launch | Dated external leaderboard claim cited by NVIDIA |
| PAI-Bench and Physics-IQ | ranked first in disclosed world-generation categories | Physical-world generation comparison |
| RoboLab | ranked first for robot policy | Robot-policy leaderboard result reported by NVIDIA |
These are release-time results, not independently reproduced guarantees. Leaderboard positions span different modalities and harnesses, and open weights do not remove the need for domain-specific physical validation. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
Cosmos 3 Super, Cosmos 3 Nano and Cosmos 3 Edge is described as mixture-of-transformers omni-model family for vision reasoning, world generation and world-action prediction with physical-scene and video context varies by checkpoint and task rather than one launch token headline of stated context. Its access position at verification time is downloadable under OpenMDW 1.1 with Hugging Face, GitHub and NVIDIA tooling. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For Cosmos 3 Super, Cosmos 3 Nano and Cosmos 3 Edge, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
Cosmos 3 is a major foundation-model family and a clear gap in the previous register. Teams should evaluate each checkpoint in the exact simulator, sensor and action stack they intend to deploy.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
- NVIDIA Cosmos 3 release overview
- NVIDIA Cosmos 3 newsroom announcement
- NVIDIA Cosmos 3 model collection
Image provenance
Hero image: NVIDIA official Cosmos 3 artwork. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
