NVIDIA · Open Research Weights (Hugging Face)

NVIDIA Nemotron Labs Dory 1.0

Ultra-low VRAM memory residency (4.4 GB FP8), 256k long-context needle retrieval, and recurrent layer depth execution.

SWE-bench Verified 66.4%
MMLU-Pro 82.8%
MATH-500 85.6%
Context Window 256,000 tokens
Specification Value
Developer / Lab NVIDIA
Parameters Total 4 Billion Parameters (Looped Transformer)
Parameters Active 4.1B Active Recurrent Block (Equivalent to 16B Depth)
Architecture Type Looped Transformer Weight Sharing + FlashAttention-4 + Adaptive Recurrent Halting
License Open Research Weights (Hugging Face)
Input Pricing (per 1M) $0.15
Output Pricing (per 1M) $0.45

Head-to-Head Comparisons with NVIDIA Nemotron Labs Dory 1.0