- Terminal Bench 2.1 56.4
- GDPVal (metric not stated) 46.7
- SWE-Bench Verified 70.7
- SWE-Bench Multilingual 67.7
- ProfBench (Search) 56
- PinchBench 90
- Mean of TauBench V3 Airline 81.5 / Retail 86.4 / Telecom 92.9 86.9
- TauBench V3 Banking 22.6
- BrowseComp 44.4
- Vals.ai Financial Agent 1.1 without web search (with web search 53.7) 60.1
- IOI 2025 570
- LiveCodeBench (v6) 89
- IMOAnswerBench (no tools); with tools 92.3 88.6
- Apex-Shortlist (no tools); with tools 84.8 74.9
- GPQA (no tools) 87
- SciCode (subtask) 44.6
- HLE (no tools); with tools 37.4 26.7
- CritPt (no tools) 3.1
- MMLU-Pro 86.8
- OmniScience Accuracy 24.1 (Non-Hallucination 78.7), not the net index 24.1
- IFBench (prompt loose) 81.7
- Multi-Challenge 63.8
- AA-LCR 65.4
- RULER (1M) 94.7
- Longbench v2 (≤ 1M) 61.9
- MMLU-ProX (avg en/de/fr/es/it/ja/zh/hi/pt/ko) 83
- WMT24++ (en→xx) 83.7
Nemotron 3 Ultra
Nemotron 3 Ultra ranks first on IMO-ProofBench.
118
Benchmarks with a score
1
Ranked first, out of five or more models
4
In the top three
NVIDIA Jun 2026 Nemotron 3 Ultra
Open weights
146.9
#84 of 256
Epoch Capabilities Index
Self-Reported Index 145.5 #128 of 404
- IFBench 81.7% #4 /80
- SciCode 44.6% #20 /49
- AA-LCR (Artificial Analysis Long Context Reasoning) 65.4% #17 /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 | |
|---|---|---|---|---|
| IFBench | 81.7% | #4 of 80 | Lab model card | |
| MMLU-Pro | 86.8% | #12 of 167 | Lab model card | |
| RULER | 94.7% | #8 of 42 | Lab model card | |
| MMLU-ProX | 83% | #5 of 23 | Lab model card | |
| LongBench v2 | 61.9% | #9 of 35 | Lab model card | |
| LMArena (Elo) | 1,426 | #95 of 335 | Official leaderboard | |
| MultiChallenge | 63.8% | #12 of 42 | Lab model card | |
| BullshitBench | 41.8% | #33 of 107 | Official leaderboard | |
| BullshitBench V2 | 49% | #53 of 129 | Official leaderboard | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 65.4% | #17 of 38 | Lab model card | |
| StudyArena Text | 1,229 | #45 of 85 | Official leaderboard | |
| Humanity's Last Exam | 26.7% | #86 of 154 | Lab model card | |
| Last Translation Benchmark | 4.83 | #22 of 36 | Official leaderboard | |
| SEA-HELM | 64.1 | #45 of 71 | Official leaderboard | |
| WMT24++ | 83.7% | #11 of 15 | Lab model card | |
| AA-LCR Extended | 41.6% | #40 of 49 | Official leaderboard | |
| DesignArena ASCII Art | 1,093 | #68 of 77 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IMO-AnswerBench | 88.6% | #7 of 37 | Lab model card | |
| IMO-ProofBench | 82.3% | #1 of 5 | Lab model card | |
| DTBench | 90.1% | #50 of 180 | Official leaderboard | |
| USAMO 2026 | 97.6% | #2 of 6 | Lab model card | |
| GPQA Diamond | 85.3% | #79 of 229 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 86.7% | #82 of 212 | Epoch AI run | |
| SimpleBench | 41.7% | #70 of 106 | Official leaderboard | |
| IMO 2025 | 83.3% | #4 of 6 | Lab model card | |
| CritPt | 3.10% | #12 of 13 | Lab model card | |
| ProofBench | 2.00% | #56 of 59 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 24.2% | #26 of 27 | Official leaderboard | |
| MathArena Apex Shortlist | 74.9% | #4 of 4 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 89% | #9 of 90 | Lab model card | |
| IOI 2025 | 570.0 | #1 of 3 | Lab model card | |
| SciCode | 44.6% | #20 of 49 | Lab model card | |
| SWE-bench Verified | 70.7% | #73 of 155 | Lab model card | |
| StructEval | 74.3% | #15 of 28 | Official leaderboard | |
| WeirdML | 43.5% | #65 of 112 | Official leaderboard | |
| SWE-bench Multilingual | 67.7% | #47 of 73 | Lab model card | |
| DesignArena 3D Design | 1,111 | #90 of 136 | Official leaderboard | |
| DesignArena Game Dev | 1,127 | #95 of 143 | Official leaderboard | |
| DesignArena Code | 1,133 | #101 of 143 | Official leaderboard | |
| DesignArena Website | 1,131 | #107 of 151 | Official leaderboard | |
| DesignArena UI Component | 1,128 | #99 of 139 | Official leaderboard | |
| DesignArena SVG | 1,061 | #81 of 103 | Official leaderboard | |
| DesignArena Data Visualization | 1,138 | #94 of 118 | Official leaderboard | |
| Vibe Code Bench | 7.64% | #104 of 120 | Official leaderboard | |
| Code Migration | 4.91% | #72 of 78 | Official leaderboard | |
| SWE Atlas v1.1 | 21.4% | #27 of 28 | Official leaderboard | |
| SWE-bench Verified Extended | 8.70% | #43 of 44 | Official leaderboard | |
| APEX-SWE | 15.2% | #64 of 65 | Official leaderboard | |
| Snorkel Agentic Coding 2.0 | 5.10% | #16 of 16 | Official leaderboard | |
| SWE-bench CLI | 0.50% | #16 of 16 | Official leaderboard | |
| Real-SWE 1.1 | 4.80% | #15 of 15 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PinchBench 2.0 | 89.9% | #4 of 57 | Official leaderboard | |
| tau3-bench | 86.9% | #2 of 14 | Lab model card | |
| PinchBench | 90% | #5 of 14 | Lab model card | |
| SnakeBench | 24 | #91 of 188 | Official leaderboard | |
| Mystery Game Puzzles | 20% | #49 of 84 | Epoch AI run | |
| τ³-bench Banking | 22.6% | #14 of 22 | Lab model card | |
| RuneBench | 3.76 | #59 of 76 | Official leaderboard | |
| MCP Atlas | 63.1% | #29 of 37 | Official leaderboard | |
| BrowseComp | 44.4% | #68 of 81 | Lab model card | |
| Terminal-Bench 2.1 | 56.4% | #48 of 56 | Lab model card | |
| Agent Arena | -15.60% | #52 of 56 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 11.8% | #41 of 44 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 30.6% | #17 of 18 | Official leaderboard | |
| Toolathlon-Verified | 34.3% | #24 of 25 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #53 of 53 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnorkelUnderwrite 2.0 | 30% | #3 of 16 | Official leaderboard | |
| CorpFin v2 | 65.5% | #28 of 119 | Official leaderboard | |
| TaxEval v2 | 73.1% | #51 of 129 | Official leaderboard | |
| ProfBench | 56% | #1 of 2 | Lab model card | |
| LegalBench | 82.1% | #74 of 138 | Official leaderboard | |
| Vals Finance Agent | 60.1% | #6 of 10 | Lab model card | |
| MedCode | 38.6% | #66 of 96 | Official leaderboard | |
| SnorkelLegal | 29.4% | #11 of 16 | Official leaderboard | |
| FoodTruck Bench | $3,517 | #43 of 59 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 0.42% | #63 of 80 | Official leaderboard | |
| SnorkelRevOps | 10.1% | #13 of 16 | Official leaderboard | |
| Legal Research Bench | 15.4% | #76 of 93 | Official leaderboard | |
| YC-Bench | $326,869 | #49 of 59 | Official leaderboard | |
| Vals Finance Agent v2 | 37.7% | #67 of 80 | Official leaderboard | |
| Vals Index v2 | 27.4% | #55 of 65 | Official leaderboard | |
| Excel Modeling Benchmark | 31.7% | #65 of 76 | Official leaderboard | |
| APEX-1 | 56.1% | #61 of 71 | Official leaderboard | |
| DAYJOB: Healthcare | 0.40% | #32 of 37 | Official leaderboard | |
| JobBench | 27.2% | #14 of 16 | Official leaderboard | |
| DeepFinance | 14.5% | #21 of 23 | Official leaderboard | |
| DAYJOB: Finance | 0% | #36 of 39 | Official leaderboard | |
| GDP.xlsx | 4.90% | #19 of 20 | Official leaderboard | |
| Tuesday Work Index | 25.3 | #42 of 44 | Official leaderboard | |
| APEX-Agents | 22.7% | #59 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 38.3% | #63 of 65 | Official leaderboard | |
| GDPval Extended | 0% | #42 of 43 | Official leaderboard | |
| APEX-Accounting | 29.4% | #48 of 49 | Official leaderboard | |
| Digital Work Index | 6.95 | #16 of 16 | Official leaderboard | |
| SnorkelFinance 2.0 | 4.30% | #16 of 16 | Official leaderboard | |
| SnorkelManufacturing | 5.40% | #16 of 16 | Official leaderboard | |
| WorkplaceAgents | 6.60% | #14 of 14 | Official leaderboard | |
| GDPval | 46.7% | #10 of 10 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ABC Bench Fragment Design | 96.2% | #5 of 27 | Official leaderboard | |
| ABC Bench Liquid-Handling Robot | 96% | #14 of 34 | Official leaderboard | |
| FORTRESS | 26.4% | #40 of 73 | Official leaderboard | |
| DistressBench | 62.1% | #38 of 42 | Official leaderboard | |
| OncologyBench | 1.33% | #28 of 30 | Official leaderboard | |
| TxBench-Antibody-Discovery | 21% | #28 of 30 | Official leaderboard | |
| TxBench-Oligo-Discovery | 25.1% | #28 of 30 | Official leaderboard | |
| BioSecBench-Refusal | 0% | #21 of 22 | Official leaderboard | |
| MetagenomicsBench | 16.3% | #25 of 26 | Official leaderboard | |
| BioSecBench-Function | 21.6% | #26 of 27 | Official leaderboard | |
| BioSecBench-Surveillance | 21.2% | #26 of 27 | Official leaderboard | |
| EpiBench | 11.3% | #26 of 27 | Official leaderboard | |
| scBench-Long | 3.03% | #26 of 27 | Official leaderboard | |
| SpatialBench-Long | 15.9% | #26 of 27 | Official leaderboard | |
| VariantBench | 17.8% | #26 of 27 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 22.7% | #29 of 30 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 51.8 | #65 of 151 | Official leaderboard | |
| LMCA | 36.9% | #73 of 147 | Official leaderboard | |
| ACCoRD | 64.6 | #76 of 152 | Official leaderboard | |
| SPEED-Bench | 5.15 | #1 of 2 | Lab model card |
Its lab's numbers put it at 145.5, 1.5 points below its ECI
The Self-Reported Index fits one score to the 23 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 143.2 to 147.7. How we compute it.
- RULER 94.7
- PinchBench 90
- LiveCodeBench v6 89
- IMO-AnswerBench 88.6
- GPQA Diamond 87
- tau3-bench 86.9
- MMLU-Pro 86.8
- WMT24++ 83.7
- MMLU-ProX 83
- IFBench 81.7
- SWE-bench Verified 70.7
- SWE-bench Multilingual 67.7
- AA-LCR (Artificial Analysis Long Context Reasoning) 65.4
- MultiChallenge 63.8
- LongBench v2 61.9
- Vals Finance Agent 60.1
- Terminal-Bench 2.1 56.4
- GDPval 46.7
- SciCode 44.6
- BrowseComp 44.4
- Humanity's Last Exam 26.7
- τ³-bench Banking 22.6
- CritPt 3.1
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| GPQA Diamond | 87% GPQA (no tools) | 85.3% Epoch AI run | +1.7 pts |
NVIDIA's models over time
NVIDIA's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
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. 5 of them aren't in the launch post.
Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning Jun 9, 2026
Capabilities
- Terminal-Bench 2.1 56.4
- GDPval 46.7
- SWE-bench Verified 70.7
- SWE-bench Multilingual 67.7
- ProfBench 56
- PinchBench 90
- tau3-bench 70.9
- τ³-bench Banking 22.6
- BrowseComp 44.4
- Vals Finance Agent 60.1
- IOI 2025 570
- LiveCodeBench v6 89
- IMO-AnswerBench 88.6
- MathArena Apex Shortlist 74.9
- GPQA Diamond 87
- SciCode 44.6
- Humanity's Last Exam 26.7
- CritPt 3.1
- MMLU-Pro 86.8
- IFBench 81.7
- MultiChallenge 63.8
- AA-LCR (Artificial Analysis Long Context Reasoning) 65.4
- RULER 94.7
- LongBench v2 61.9
- MMLU-ProX 83
- WMT24++ 83.7
- IMO-ProofBench 82.3
- IMO 2025 83.3
- Putnam 2025 96.7
- USAMO 2026 97.6
Hallucination
Other
- SPEED-Bench 5.15