- MMLU Pro 81.9
- AA-Omniscience (metric not stated) 17.5
- GPQA Diamond (no tools) 75.4
- HLE (text-only, no tools) 11.7
- SciCode 32.6
- SWE-bench Verified 51.6
- SWE-bench Multilingual 39.3
- Terminal-Bench 2.1 24.6
- PinchBench 85.4
- BrowseComp 37
- τ³-bench (Banking) 9.28
- GDPval-AA-V2 832
- IFBench (loose) 71.9
- AA-LCR 52
Nemotron 3.5 Lightning 30B-A3B
Nemotron 3.5 Lightning 30B-A3B.
60
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
NVIDIA Aug 2026 Nemotron 3.5 Lightning 30B-A3B
Open weights Small
135.3
#227 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- IFBench 72% #37 /80
- SciCode 32.7% #41 /49
- AA-LCR (Artificial Analysis Long Context Reasoning) 52.4% #31 /38
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLU-Pro | 81.9% | #60 of 167 | Lab model card | |
| RULER | 91.5% | #17 of 42 | Lab model card | |
| IFBench | 72% | #37 of 80 | Lab model card | |
| LMArena (Elo) | 1,351 | #177 of 335 | Official leaderboard | |
| Creative Writing v3 | 58.0 | #89 of 120 | Official leaderboard | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 52.4% | #31 of 38 | Lab model card | |
| Humanity's Last Exam | 11.7% | #133 of 154 | Lab model card | |
| StudyArena Text | 1,180 | #76 of 85 | Official leaderboard | |
| SEA-HELM | 46.1 | #64 of 71 | Official leaderboard | |
| NV Chat Benchmark | 12.1 | #2 of 2 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 76.5% | #138 of 302 | Lab model card | |
| DTBench | 70.4% | #113 of 180 | Official leaderboard | |
| ProofBench v1.1 | 0% | #52 of 52 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 14.4% | #27 of 27 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 22.9% | #26 of 26 | Official leaderboard | |
| CritPt | 0% | #13 of 13 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 52.2% | #126 of 155 | Lab model card | |
| ALE-Bench | 1,017 | #104 of 126 | Official leaderboard | |
| SciCode | 32.7% | #41 of 49 | Lab model card | |
| Code Migration | 2.95% | #75 of 78 | Official leaderboard | |
| SWE-bench Multilingual | 39.3% | #71 of 73 | Lab model card | |
| Vibe Code Bench | 0% | #120 of 120 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PinchBench 2.0 | 86.4% | #11 of 57 | Official leaderboard | |
| PinchBench | 85.4% | #9 of 14 | Lab model card | |
| BrowseComp | 39.5% | #72 of 81 | Lab model card | |
| τ³-bench Banking | 9.90% | #20 of 22 | Lab model card | |
| Terminal-Bench 2.1 | 24.6% | #55 of 56 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CorpBench Work | 79.1 | #9 of 17 | Official leaderboard | |
| CorpFin v2 | 54.2% | #87 of 119 | Official leaderboard | |
| Excel Modeling Benchmark | 17% | #72 of 76 | Official leaderboard | |
| Legal Research Bench | 2.88% | #90 of 93 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 0% | #78 of 80 | Official leaderboard | |
| Vals Finance Agent v2 | 18.5% | #78 of 80 | Official leaderboard | |
| Tax Agent Bench | 2.59% | #83 of 84 | Official leaderboard | |
| MedScribe | 4.27% | #97 of 97 | Official leaderboard | |
| Vals Index v2 | 11.5% | #65 of 65 | Official leaderboard | |
| GDPval-AA v2 | 832 | #27 of 27 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Permit | 99.7% | #25 of 69 | Official leaderboard | |
| nocot-bench (NCRI) | 56.1 | #150 of 185 | Official leaderboard | |
| World-Class Bio | 30.9% | #62 of 74 | Official leaderboard | |
| BioTIER Refuse | 21.1% | #58 of 69 | Official leaderboard | |
| DistressBench | 63.5% | #37 of 42 | Official leaderboard | |
| Human Pathogen Capabilities Test | 26.5% | #81 of 86 | Official leaderboard | |
| Molecular Biology Capabilities Test | 26.1% | #87 of 90 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 29.5% | #43 of 44 | Official leaderboard | |
| OncologyBench | 0% | #30 of 30 | Official leaderboard | |
| TxBench-Antibody-Discovery | 16% | #30 of 30 | Official leaderboard | |
| TxBench-Oligo-Discovery | 8.55% | #30 of 30 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 10.3% | #30 of 30 | Official leaderboard | |
| BioSecBench-Function | 13.8% | #27 of 27 | Official leaderboard | |
| BioSecBench-Surveillance | 14.1% | #27 of 27 | Official leaderboard | |
| EpiBench | 6.92% | #27 of 27 | Official leaderboard | |
| scBench-Long | 0% | #27 of 27 | Official leaderboard | |
| SpatialBench-Long | 4.76% | #27 of 27 | Official leaderboard | |
| VariantBench | 14.4% | #27 of 27 | Official leaderboard | |
| MetagenomicsBench | 7.67% | #26 of 26 | Official leaderboard | |
| BioSecBench-Refusal | 0% | #22 of 22 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 54.7 | #108 of 152 | Official leaderboard | |
| Conceptual Reasoning Index | 33.6 | #110 of 151 | Official leaderboard | |
| LMCA | 20.9% | #111 of 147 | Official leaderboard |
Its lab's numbers put it at 135.3, number 227 of 404
The Self-Reported Index fits one score to the 14 capability results NVIDIA reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 133.2 to 137.4. How we compute it.
- RULER 91.5
- PinchBench 85.4
- MMLU-Pro 81.9
- GPQA Diamond 76.5
- IFBench 72
- AA-LCR (Artificial Analysis Long Context Reasoning) 52.4
- SWE-bench Verified 52.2
- BrowseComp 39.5
- SWE-bench Multilingual 39.3
- SciCode 32.7
- Terminal-Bench 2.1 24.6
- Humanity's Last Exam 11.7
- τ³-bench Banking 9.9
- CritPt 0
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 4 of them aren't in the launch post.
Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AI Oct 15, 2026
Capabilities
- SWE-bench Verified 52.2
- SWE-bench Multilingual 37.1
- tau2-bench
- τ³-bench Banking 9.9
- Terminal-Bench 2.1 24.4
- GDPval-AA v2
- PinchBench 83.9
- BrowseComp 39.5
- SciCode 32.7
- GPQA Diamond 76.5
- Humanity's Last Exam 10.8
- CritPt 0
- AA-LCR (Artificial Analysis Long Context Reasoning) 52.4
- RULER 91.5
- IFBench 72
- NV Chat Benchmark 12.1