Cosmos 3 (Nano / Super)
World model
nvidia/Cosmos3-Nano
Price not published
Cosmos Reason 2
Model idnvidia/Cosmos-Reason2-8B
PricePrice not published
Checked29 September 2026
No published training cutoff
At least1,962events since its release on 19 December 2025
85 historic, 449 major. 329 days of news.
NVIDIA doesn’t publish a training cutoff, so we count from its release date. The real gap is larger. Coverage before 2026 is thin (80 entries for 2025), so this count is a floor.
How to call it
2 ways to use Cosmos Reason 2, with the model id each provider expects.
| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face | nvidia/Cosmos-Reason2-8B |
huggingface.co | |
| Hugging Face | nvidia/Cosmos-Reason2-2B |
huggingface.co |
Pricing
Price not published.
No price source linked. Checked 29 September 2026.
Specs
- Input
- Text, images and video
- Output
- Text
- Context window
- 256,000 tokens
- Max output
- Not published
- Open weights
- Yes
- Licence
- NVIDIA Open Model License (commercial use allowed; HF repo gated - accept terms)
- Training cutoff
- Not published
- Released
- 19 December 2025
- Status
- Current
- Type
- Multimodal model
What stands out
What its maker marketed at launch, and what people found later, each with its source.
Physical-AI reasoning VLM
Spatio-temporal video reasoning, 2D/3D point and box localization, robot planning; 8B beats base Qwen3-VL-8B on robotics (56.90 vs 53.08) and self-driving (67.85 vs 46.38) evals per model card.
Backbone for GR00T N1.7
Found after launch
Cosmos-Reason2-2B is the VLM backbone of Isaac GR00T N1.7.
Notes
Based on Qwen3-VL (8B variant from Qwen3-VL-8B-Instruct, 8.7B params, 32 GB+ GPU). Initial release 2025-12-19, updated 2026-03-10; promoted at CES 2026. Up to 256K input tokens. Its role is folded into Cosmos 3 for new projects.
Sources: Cosmos-Reason2-8B card, NVIDIA forum: CES 2026 Cosmos announcements.