Atlas

Models

← All models

Llama 3.1 Nemotron Nano 4B V1.1

NVIDIA · 2025-05-20 · 4.5B parameters

Llama 3.1 Nemotron Nano 4B V1.1 is NVIDIA's open-weight, dense 4B-class reasoning model derived from Llama 3.1 8B through NVIDIA's Minitron compression process; the released checkpoint contains exactly 4,512,746,496 parameters. Released on May 20, 2025, it was post-trained for reasoning, chat, math, code, RAG, and tool calling, supports switchable reasoning through the system prompt, fits on a single RTX GPU, and has a 131,072-token context window.

Benchmark scores

BenchmarkScore
AA-LCR0.0
AIME 202550.0
Artificial Analysis Intelligence Index8.5
Artificial Analysis Openness Index52.8
Capability122.5
GPQA Diamond40.8
Humanity's Last Exam (Text-Only)5.1
IFBench (Artificial Analysis)25.5
LiveCodeBench (Artificial Analysis)49.3
MATH-500 (Artificial Analysis source)94.7
SciCode (Artificial Analysis)10.1
τ²-bench (Telecom)11.7
Loading Atlas data…