Nemotron-Labs-Sigma
NVIDIA released Nemotron-Labs-Sigma, a family of 14 open checkpoints at 3B and 8B that form a continuous diffusion language model working in a 16-D embedding space.
Most diffusion language models such as LLaDA and Dream work on discrete tokens. Sigma denoises text in a smooth 16-dimensional latent space instead. NVIDIA says this makes few-step distillation and trajectory steering practical for text.
- Date
- Friday 20 November 2026
- Lab
- NVIDIA
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Checkpoints | 14 3B and 8B sizes | company |
| Context length | 4,096 tokens 8,192 for Sigma-8B-instruct | company |
Based on a MarkTechPost article only. The license and benchmark results were not visible in the text I read.
Sources
This record was checked against its sources on 21 November 2026. How we check