Nemotron-Labs-Diffusion Models by NVIDIA AI
Last refreshed 2026-06-20. Next refresh: weekly.
Details
About
NVIDIA Nemotron-Labs-Diffusion is a family of diffusion language models (DLMs) released by NVIDIA Research in May 2026. Unlike traditional autoregressive models, they generate text by producing multiple tokens in parallel and iteratively refining them, enabling up to 6× throughput over comparable AR models. Available in 3B, 8B, and 14B sizes with base and instruct variants; an 8B VLM is also included.
Decision facts
- Best fit
- coding
- Capability starting point
- Nemotron-Labs-Diffusion 14B with 128K context
- Lowest tracked input
- Not tracked
- Closest related family
- NemoGuard
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs 128K context, 14B parameters, and fine tuning.
Use when the workload needs 128K context and 3B parameters.
Use when the workload needs 128K context and 8B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Nemotron-Labs-Diffusion 14B | Use when the workload needs 128K context, 14B parameters, and fine tuning. | 2026-05 | 128K context14B parametersfine tuning | Current |
| Nemotron-Labs-Diffusion 3B | Use when the workload needs 128K context and 3B parameters. | 2026-05 | 128K context3B parameters | Current |
| Nemotron-Labs-Diffusion 8B | Use when the workload needs 128K context and 8B parameters. | 2026-05 | 128K context8B parameters | Current |
Release Timeline
1 release groupSpecifications(3 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Nemotron-Labs-Diffusion 14B | 2026-05 | 128K | 14B |
| Nemotron-Labs-Diffusion 3B | 2026-05 | 128K | 3B |
| Nemotron-Labs-Diffusion 8B | 2026-05 | 128K | 8B |
