Skip to content
AI Model Radar

NVIDIA Nemotron Labs Dory1.0 4B BF16

nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16
Open Source · chat · open-weights
Open Alert me on changes
Context
262.1K
Max output
—
Weights
Open
API $/1M
—
Modalities
text
Released
16 Oct 2026
License: other · nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16
Download image Share on X Share on LinkedIn
AI summary
● machine-written

NVIDIA Nemotron Labs Dory 1.0 4B: Open-weights model with 256K context

A 4-billion-parameter open-weights chat model from NVIDIA pre-trained on 6 trillion tokens with 256K context length support. The model uses a latent-recurrent architecture with adjustable inference loops (1, 2, or 4 passes) to trade computation for quality. Available in multiple quantization formats and released under OpenMDW-1.1 license for both commercial and non-commercial use.

What's new
  • Pre-trained on 6 trillion tokens using mixed NVFP4 and BF16 precision, followed by 300B-token context-extension phase
  • Supports 256K context length with adjustable latent-recurrent inference loops
  • Architecture: eight-layer input stack, 20-layer recurrent block, eight-layer output stack
  • Available in BF16, MXFP8, NVFP4, and Q5_K_M GGUF quantization formats
  • Released under OpenMDW-1.1 license supporting commercial and non-commercial use
Best for
Long-context text understanding and generation tasksEfficient deployment in resource-constrained environmentsAdjustable inference compute budgets without model switching
Sources

Source: https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16