Nemotron-Labs TwoTower Models by NVIDIA AI
1 model2026Up to 128k ctx
Last refreshed 2026-07-02. Next refresh: weekly.
Details
ResearcherNVIDIA AI
LicenseNVIDIA Open Model
Commercial useCommercial use: permitted
Models1
Released2026
Max context128k
About
Nemotron-Labs TwoTower is NVIDIA's open-weight block-wise autoregressive diffusion language model family that adapts a Nemotron-3-Nano-30B-A3B autoregressive backbone into a two-tower architecture with a frozen AR/context tower and a trainable diffusion/denoiser tower.
Decision facts
- Best fit
- General model comparison
- Capability starting point
- Nemotron-Labs TwoTower 30B-A3B Base with 128k context
- Lowest tracked input
- Not tracked
- Closest related family
- NVIDIA Nemotron Nano 12B v2 VL
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
1 in view
Use when the workload needs 128k context and open source.
2026-06128k contextopen source
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Nemotron-Labs TwoTower 30B-A3B Base | Use when the workload needs 128k context and open source. | 2026-06 | 128k contextopen source | Current |
Release Timeline
1 release group2026-06
1 current
Nemotron-Labs TwoTower 30B-A3B Base
Current128k contextopen source
Specifications(1 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Nemotron-Labs TwoTower 30B-A3B Base | 2026-06 | 128k | ~60B total checkpoint; Hugging Face reports 63B params |
