Nemotron-Labs-Dory Models by NVIDIA AI
NVIDIA AI releases · 26 in the last 12 months · this family litChangelog →
1 model2026Up to 256K ctx
Last refreshed 2026-10-20. Next refresh: weekly.
Details
ResearcherNVIDIA AI
LicenseOpenMDW 1.1
Commercial usePermitted
Models1
Released2026
Max context256K
Capabilities
ReasoningAll models
JSON / Tool useAll models
About
NVIDIA Nemotron-Labs-Dory is an NVIDIA research family of looped (latent-recurrent) normalized-GPT language models. Dory 1.0 4B, released October 16, 2026, re-runs its recurrent layers 1 to 4 times per token so it can spend more compute per token without adding parameters, and supports last-loop KV caching. Weights ship in BF16, MXFP8, NVFP4 and GGUF under OpenMDW-1.1.
Decision facts
- Best fit
- reasoningJSON / Tool use
- Capability starting point
- Nemotron Labs Dory 1.0 4B with 256K context and reasoning and JSON / Tool use
- 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.
1 in view
Nemotron Labs Dory 1.0 4BCurrent
Use when the workload needs 256K context, 4B parameters, and reasoning.
2026-10256K context4B parametersreasoning
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Nemotron Labs Dory 1.0 4B | Use when the workload needs 256K context, 4B parameters, and reasoning. | 2026-10 | 256K context4B parametersreasoning | Current |
Release Timeline
1 release group2026-10
1 current
Nemotron Labs Dory 1.0 4B
Current256K context4B parametersreasoning
Specifications(1 models)
| Model | Released | Context | Parameters | Reasoning | JSON / Tool use |
|---|---|---|---|---|---|
| Nemotron Labs Dory 1.0 4B | 2026-10 | 256K | 4B | Yes | Yes |
