- GPQA-Diamond 89.5
- HLE, text-only (no tools); HLE full set 29.20 31.8
- MMLU-ProX 76.9
- AA-Omniscience index 4.03 (accuracy 26.20, hallucination rate 30.0) 4.03
- CritPt 10.9
- APEX shortlist 88.4
- LiveCodeBench v6 88.2
- SciCode, subtask 47.1
- AA-LCR v1.1 68.6
- GraphWalks 89.7
- IFBench 70.9
- LMArena v3 proxy (NV Chat Benchmark in the technical report) 52.7
- WMT24++ (COMET) 84.8
- Mean of tau2-bench Retail 69.24 / Airline 76.50 / Telecom 73.19 73
- tau3-bench banking 18.9
- GDPval-AA v2 1211
- BrowseComp (78.91 in the technical report) 78
- SWE-bench Verified (OpenCode harness) 71.1
- SWE-bench Multilingual (OpenCode harness) 67.7
- SWE-bench Pro (OpenCode harness) 51.8
- Terminal-Bench 2.1 (Terminus 2) 60.7
- BabyVision 29.6
- MMLongBench-Doc 62.8
- V* 85.6
- OCR-Reasoning 65.1
- MathVision 78.2
- OCRBench v2, EN 68.12 / ZH 59.15 68.1
- CharXiv reasoning 80.8
- Video-MME v2 (subtitle setting not stated) 32.9
- OSWorld-Verified 51.1
- WebVoyager [offline] 77.7
- AutomationBench-AA [offline] 32.4
- GDP.pdf [offline] 6.6 (AA 6.6) 6.6
- Terminal Bench 4 [offline] 1
- DeepSWE [offline] 24.6
- AA-Briefcase-Lite offline, Nemotron 3.5 Super 1137
Nemotron 3.5 Super VL
Nemotron 3.5 Super VL ranks first on 3 of the 85 benchmarks with five or more models scored.
101
Benchmarks with a score
3
Ranked first, out of five or more models
5
In the top three
NVIDIA Oct 2026 Nemotron 3.5 Super VL
Open weights Flagship
144.1
#141 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- IFBench 70.9% #42 /80
- SciCode 47.1% #16 /49
- Terminal-Bench 4.0 0.30% #39 /40
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 | |
|---|---|---|---|---|
| GraphWalks | 89.7% | #8 of 27 | Lab model card | |
| LMArena (Elo) | 1,415 | #111 of 335 | Official leaderboard | |
| WMT24++ | 84.8% | #6 of 15 | Lab model card | |
| Humanity's Last Exam | 31.8% | #76 of 154 | Lab model card | |
| NV Chat Benchmark | 52.7 | #1 of 2 | Lab model card | |
| IFBench | 70.9% | #42 of 80 | Lab model card | |
| MMLU-ProX | 76.9% | #14 of 23 | Lab model card | |
| AA-LCR 1.1 | 68.6% | #5 of 6 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 89.5% | #37 of 302 | Lab model card | |
| MathArena Apex Shortlist | 88.4% | #2 of 4 | Lab model card | |
| CritPt | 10.9% | #9 of 13 | Lab model card | |
| ARC-AGI-2 | 1.94% | #78 of 98 | Official leaderboard | |
| ARC-AGI | 22.7% | #79 of 95 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 88.2% | #10 of 90 | Lab model card | |
| Bird-SQL | 59.6% | #2 of 13 | Lab model card | |
| SciCode | 47.1% | #16 of 49 | Lab model card | |
| SWE-bench Verified | 71.1% | #70 of 155 | Lab model card | |
| ALE-Bench | 1,352 | #69 of 126 | Official leaderboard | |
| DesignArena 3D Design | 1,117 | #85 of 136 | Official leaderboard | |
| SWE-bench Multilingual | 67.7% | #46 of 73 | Lab model card | |
| DesignArena UI Component | 1,154 | #89 of 139 | Official leaderboard | |
| DesignArena Website | 1,160 | #97 of 151 | Official leaderboard | |
| Code Arena | 1,401 | #84 of 128 | Official leaderboard | |
| DesignArena Code | 1,140 | #99 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,077 | #108 of 143 | Official leaderboard | |
| SWE-Bench Pro | 51.8% | #63 of 81 | Lab model card | |
| Vibe Code Bench | 3.71% | #108 of 120 | Official leaderboard | |
| DesignArena Data Visualization | 1,085 | #108 of 118 | Official leaderboard | |
| APEX-SWE | 22.3% | #60 of 65 | Official leaderboard | |
| Image-to-WebDev Arena | 1,392 | #51 of 55 | Official leaderboard | |
| SWE Atlas v1.1 | 24.8% | #26 of 28 | Official leaderboard | |
| Code Migration v1.1 | 16.4% | #28 of 30 | Official leaderboard | |
| DeepSWE | 24.6% | #56 of 56 | Lab model card | |
| DesignArena Image to HTML | 1,044 | #51 of 51 | Official leaderboard | |
| DesignArena Video to Website | 1,005 | #19 of 19 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseComp | 78.9% | #31 of 81 | Lab model card | |
| AutomationBench | 32.4% | #29 of 47 | Lab model card | |
| RuneBench | 4.36 | #48 of 76 | Official leaderboard | |
| tau2-bench | 73% | #52 of 79 | Lab model card | |
| τ³-bench Banking | 18.9% | #16 of 22 | Lab model card | |
| Terminal-Bench 2.1 | 60.7% | #45 of 56 | Lab model card | |
| WebVoyager | 77.7% | #10 of 12 | Lab model card | |
| OSWorld-Verified | 52.1% | #35 of 41 | Lab model card | |
| DuelLab GameBench 2 | 11.2 | #53 of 57 | Official leaderboard | |
| Agent Arena | -15.99% | #53 of 56 | Official leaderboard | |
| ScreenSpot-Pro | 43.5% | #31 of 32 | Lab model card | |
| Terminal-Bench 4.0 | 0.30% | #39 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 0.30% | #42 of 43 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Tax Agent Bench | 11.6% | #67 of 84 | Official leaderboard | |
| APEX-1 | 57.6% | #57 of 71 | Official leaderboard | |
| GDPval-AA v2 | 1,211 | #23 of 27 | Lab model card | |
| AA-Briefcase | 1,137 | #6 of 7 | Lab model card | |
| Legal Research Bench | 10.1% | #84 of 93 | Official leaderboard | |
| APEX-Accounting | 38.1% | #45 of 49 | Official leaderboard | |
| APEX-Agents 1.1 | 45.9% | #61 of 65 | Official leaderboard | |
| Vals Index v2.2 | 29.8% | #35 of 37 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 3.75% | #36 of 38 | Official leaderboard | |
| Vals Finance Index | 41% | #38 of 40 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 52.8% | #40 of 42 | Official leaderboard | |
| Vals Legal Index | 55.4% | #37 of 38 | Official leaderboard | |
| Vals Finance Agent v2.1 | 47.7% | #39 of 40 | Official leaderboard | |
| RWE-bench | 0% | #15 of 15 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLongBench-Doc | 66.6% | #1 of 17 | Official leaderboard | |
| OCRBench v2 | 67.4% | #2 of 14 | Official leaderboard | |
| MMSI-Bench | 43.1% | #1 of 6 | Lab model card | |
| OmniSpatial | 58.6% | #1 of 6 | Lab model card | |
| BLINK | 71.3% | #6 of 29 | Lab model card | |
| OCR-Reasoning | 65.1% | #1 of 3 | Lab model card | |
| CharXiv | 80.8% | #31 of 75 | Lab model card | |
| MATH-Vision (MathVision) | 78.2% | #24 of 51 | Lab model card | |
| Vision Arena | 1,212 | #72 of 136 | Official leaderboard | |
| ERQA | 57.5% | #21 of 36 | Lab model card | |
| ViewSpatial-Bench | 52% | #2 of 3 | Lab model card | |
| Video-MME | 74.4% | #38 of 51 | Lab model card | |
| MindCube | 62.2% | #3 of 4 | Lab model card | |
| V* | 85.6% | #13 of 17 | Lab model card | |
| Humanity's Sixth Sense | 24.7% | #29 of 34 | Official leaderboard | |
| VSI-Bench | 48.1% | #6 of 7 | Lab model card | |
| BabyVision | 29.6% | #23 of 26 | Lab model card | |
| BVB | 52.1 | #43 of 45 | Official leaderboard | |
| GDP.pdf | 6.60% | #17 of 17 | Lab model card | |
| CVBench | 69.6% | #15 of 15 | Lab model card | |
| VideoMME v2 (with subtitles) | 32.9% | #6 of 6 | Lab model card | |
| PAI-Bench | 63.8% | #2 of 2 | Lab model card | |
| DSI-Bench | 61.4% | #1 of 1 | Lab model card | |
| MMSI-Video-Bench | 34.5% | #1 of 1 | Lab model card | |
| Omni3D | 53.4% | #1 of 1 | Lab model card | |
| OSI-Bench | 30.6% | #1 of 1 | Lab model card | |
| PerceptionComp | 39.4% | #1 of 1 | Lab model card | |
| SPAR-Bench | 64.2% | #1 of 1 | Lab model card | |
| SpatialTree | 58.1% | #1 of 1 | Lab model card | |
| SPBench | 63.8% | #1 of 1 | Lab model card | |
| VSTI-Bench | 67.3% | #1 of 1 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Permit | 99.9% | #11 of 69 | Official leaderboard | |
| Molecular Biology Capabilities Test | 52.5% | #30 of 90 | Official leaderboard | |
| Human Pathogen Capabilities Test | 52.4% | #43 of 86 | Official leaderboard | |
| World-Class Bio | 43.4% | #44 of 74 | Official leaderboard | |
| BioTIER Refuse | 32.1% | #46 of 69 | Official leaderboard | |
| VCT-v2 | 26% | #24 of 25 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 47% | #28 of 31 | Official leaderboard | |
| SPEED-Bench | 4.72 | #2 of 2 | Lab model card |
Its lab's numbers put it at 144.1, number 141 of 404
The Self-Reported Index fits one score to the 32 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 141.8 to 146.3. How we compute it.
- GraphWalks 89.7
- GPQA Diamond 89.5
- LiveCodeBench v6 88.2
- V* 85.6
- WMT24++ 84.8
- CharXiv 80.8
- BrowseComp 78.9
- MATH-Vision (MathVision) 78.2
- MMLU-ProX 76.9
- Video-MME 74.4
- tau2-bench 73
- BLINK 71.3
- SWE-bench Verified 71.1
- IFBench 70.9
- CVBench 69.6
- OCRBench v2 68.1
- SWE-bench Multilingual 67.7
- MMLongBench-Doc 62.8
- Terminal-Bench 2.1 60.7
- Bird-SQL 59.6
- OSWorld-Verified 52.1
- SWE-Bench Pro 51.8
- SciCode 47.1
- ScreenSpot-Pro 43.5
- AutomationBench 32.4
- Humanity's Last Exam 31.8
- BabyVision 29.6
- DeepSWE 24.6
- τ³-bench Banking 18.9
- CritPt 10.9
- GDP.pdf 6.6
- Terminal-Bench 4.0 1
Its lab's numbers next to 4 leaderboards
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 | What differs |
|---|---|---|---|---|
| MMLongBench-Doc | 62.8% MMLongBench-Doc | 66.6% Leaderboard, VL; 12B activated (120B total) | −3.8 pts | |
| OCRBench v2 | 68.1% OCRBench v2, EN 68.12 / ZH 59.15 | 67.4% Leaderboard | +0.7 pts | |
| Terminal-Bench 4.0 | 1.00% Terminal Bench 4 [offline] | 0.30% Leaderboard, none, mini-swe-agent | +0.7 pts | |
| LMArena (Elo) | 1,420 Text Arena, Overall, With Style Control, 1420 +/- 12, 2,328 votes, preliminary NVFP4 (Table 11) | 1,415 Leaderboard, nvfp4 | +5 | NVIDIA reports a preliminary Arena rating from 2,328 votes, while the board reports a later rating. |
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. 23 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 71.1
- SWE-bench Multilingual 67.7
- SWE-Bench Pro 51.8
- tau2-bench
- τ³-bench Banking 18.9
- Terminal-Bench 2.1 60.7
- GDPval-AA v2 1211.2
- BrowseComp 78.9
- SciCode 47.1
- LiveCodeBench v6 88.2
- GPQA Diamond 89.5
- CritPt 10.9
- Humanity's Last Exam 31.8
- MathArena Apex Shortlist 88.4
- BabyVision 29.6
- CharXiv 80.8
- MATH-Vision (MathVision) 78.2
- OCR-Reasoning 65.1
- OCRBench v2 68.1
- VideoMME v2 (with subtitles) 32.9
- V* 85.6
- MMLongBench-Doc 62.8
- OSWorld-Verified 52.1
- AA-LCR 1.1 68.6
- GraphWalks 89.7
- IFBench 70.9
- NV Chat Benchmark 52.7
- MMLU-ProX 76.9
- WMT24++ 84.8
- Bird-SQL 59.6
- ScreenSpot-Pro 43.5
- LMArena (Elo) 1420
- ERQA 57.5
- Omni3D 53.4
- OmniSpatial 58.6
- SPBench 63.8
- MindCube 62.2
- MMSI-Bench 43.1
- SPAR-Bench 64.2
- BLINK 71.3
- SpatialTree 58.1
- ViewSpatial-Bench 52
- OSI-Bench 30.6
- PAI-Bench 63.8
- VSI-Bench 48.1
- DSI-Bench 61.4
- CVBench 69.6
- PerceptionComp 39.4
- Video-MME 74.4
- SPEED-Bench 4.72
- VSTI-Bench 67.3
- MMSI-Video-Bench 34.5