Atlas

Models

← All models

Nemotron 3.5 Lightning 30B A3B (precision unspecified)

NVIDIA · 2026-08-11

NVIDIA Nemotron 3.5 Lightning labels whose evaluators do not identify the served BF16 or NVFP4 artifact. Both releases support optional thinking and up to 1,048,576-token context under OpenMDW-1.1. This benchmark entry preserves the source observations without assigning them to either checkpoint or including the unresolved identity in the capability fit.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo509
AA-Briefcase v1.1 Elo500
AA-LCR v1.160.3
AA-Omniscience Index-17.7
ALE-Bench1017
Antidote: Everyday Edition888
Artificial Analysis Intelligence Index v4.313.6
Artificial Analysis Intelligence Index v4.3.212.9
Artificial Analysis Omniscience Accuracy14.4
Artificial Analysis Omniscience Hallucination Rate37.6
Artificial Analysis Openness Index83.3
AutomationBench-AA 1.0.60.8
BuseyBench SVG2.1
ComplexConstraints7.1
CritPt (avg@5)0.0
EnterpriseBench: CoreCraft10.3
EnterpriseOps-Gym-AA18.4
GDP.pdf (Artificial Analysis, All-pass)3.0
GDP.pdf (Artificial Analysis, Mean Criteria)46.5
GDPval-AA v2767
GDPval-AA v2.1624
GPQA Diamond (Artificial Analysis avg@5)74.3
HANDBOOK.md0.0
Hemingway-bench888
Humanity's Last Exam (Text-Only)10.6
Medical Long Context Reasoning (MLCR-AA)1.1
MLCR-AA Accuracy (Judged Responses)37.1
MLCR-AA Completeness (Judged Responses)4.6
MLCR-AA Conciseness97.2
SciCode 1.0.1 (Artificial Analysis)32.1
Loading Atlas data…