GPT-5.4 Nano ranks first on Vectara Hallucination Leaderboard (2025 dataset).
Closed weights
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vectara Hallucination Leaderboard (2025 dataset) | 3.10% | #1 of 104 | Official leaderboard | |
| Global PIQA | 85.4% | #6 of 44 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 27733.0 | #26 of 77 | Official leaderboard | |
| LMArena (Elo) | 1,401 | #130 of 335 | Official leaderboard | |
| Humanity's Last Exam | 37.7% | #67 of 154 | Lab model card | |
| Longform Creative Writing (EQ-Bench) | 57.8 | #70 of 119 | Official leaderboard | |
| GraphWalks | 73.4% | #17 of 27 | Lab model card | |
| Judgemark v2.1 | 64.5 | #27 of 42 | Official leaderboard | |
| HELM Arabic Enterprise | 63.2 | #15 of 23 | Official leaderboard | |
| OpenAI MRCR v2 | 33% | #32 of 47 | Official leaderboard | |
| STALE | 9.80% | #7 of 10 | Official leaderboard | |
| BullshitBench | 16.4% | #79 of 107 | Official leaderboard | |
| BullshitBench V2 | 14% | #106 of 129 | Official leaderboard | |
| HELM Arabic | 68.2 | #25 of 28 | Official leaderboard | |
| Judgemark v4 | 27.5 | #37 of 40 | Official leaderboard | |
| SimpleQA Verified | 11.7% | #91 of 94 | Epoch AI run |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | #77 of 212 | Epoch AI run | |
| FrontierMath | 25.9% | #20 of 55 | Official leaderboard | |
| GPQA Diamond | 78.5% | #101 of 229 | Epoch AI run | |
| FrontierMath Tier 4 | 6.25% | #25 of 55 | Epoch AI run | |
| DTBench | 80.3% | #91 of 180 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 44.9% | #57 of 86 | Official leaderboard | |
| ARC-AGI-2 | 5.69% | #66 of 98 | Official leaderboard | |
| ARC-AGI | 51.5% | #67 of 95 | Official leaderboard | |
| BixBench3 | 27.5 | #16 of 20 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 12.2% | #55 of 68 | Official leaderboard | |
| ProofBench | 5.00% | #51 of 59 | Official leaderboard | |
| ForecastBench | 0.17 | #79 of 82 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WeirdML | 49.2% | #48 of 112 | Official leaderboard | |
| IOI (Vals) | 15.3% | #29 of 55 | Official leaderboard | |
| ALE-Bench | 1,312 | #72 of 126 | Official leaderboard | |
| Vibe Code Bench | 26.1% | #80 of 120 | Official leaderboard | |
| Code Migration | 14.5% | #56 of 78 | Official leaderboard | |
| StructEval | 69.8% | #21 of 28 | Official leaderboard | |
| SWE-Bench Pro | 52.4% | #61 of 81 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 92.5% | #11 of 79 | Lab model card | |
| TextQuests | 15.2 | #7 of 28 | Official leaderboard | |
| Chess Puzzles | 45% | #29 of 93 | Epoch AI run | |
| LisanBench | 1426.0 | #49 of 133 | Official leaderboard | |
| LLM Chess | 207 | #53 of 123 | Official leaderboard | |
| PinchBench | 69.5% | #23 of 48 | Official leaderboard | |
| Toolathlon | 35.5% | #28 of 50 | Official leaderboard | |
| PinchBench 2.0 | 69% | #34 of 57 | Official leaderboard | |
| Terminal-Bench 2.0 | 46.3% | #44 of 63 | Lab model card | |
| Mystery Game Puzzles | 9.00% | #73 of 84 | Epoch AI run | |
| RuneBench | 2.35 | #69 of 76 | Official leaderboard | |
| MCP Atlas | 56.1% | #43 of 46 | Lab model card | |
| SnakeBench | 13 | #178 of 188 | Official leaderboard | |
| OSWorld-Verified | 39% | #39 of 41 | Lab model card | |
| o11y-bench | 10.9% | #22 of 22 | Official leaderboard | |
| CAR-bench | 36.7% | #12 of 12 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CorpFin v2 | 61.2% | #57 of 119 | Official leaderboard | |
| MedCode | 41% | #56 of 96 | Official leaderboard | |
| Vals Finance Agent | 47.8% | #30 of 49 | Official leaderboard | |
| Vals Index v1 | 51.9% | #21 of 34 | Official leaderboard | |
| MedScribe | 77.1% | #65 of 97 | Official leaderboard | |
| LegalBench | 77.9% | #97 of 138 | Official leaderboard | |
| MortgageTax | 59.1% | #64 of 87 | Official leaderboard | |
| TaxEval v2 | 67.4% | #98 of 129 | Official leaderboard | |
| Vals Index v2 | 33% | #50 of 65 | Official leaderboard | |
| Excel Modeling Benchmark | 44.8% | #59 of 76 | Official leaderboard | |
| Vals Finance Agent v2 | 38.2% | #64 of 80 | Official leaderboard | |
| CaseLaw v2 | 51.9% | #46 of 53 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 0% | #71 of 80 | Official leaderboard | |
| Legal Research Bench | 6.25% | #87 of 93 | Official leaderboard | |
| Tax Agent Bench | 4.97% | #80 of 84 | Official leaderboard | |
| YC-Bench | $39,388 | #57 of 59 | Official leaderboard | |
| PolicyBench | 65.3% | #50 of 50 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU-Pro | 69.5% | #51 of 90 | Lab model card | |
| Vision Arena | 1,199 | #81 of 136 | Official leaderboard | |
| MineBench | 1,248 | #52 of 75 | Official leaderboard | |
| ParseBench | 42.2% | #35 of 44 | Official leaderboard | |
| GeoGuesser environment (FineEnvs) | 0.26 | #12 of 14 | Official leaderboard | |
| vCX Benchmark | 74.5% | #39 of 42 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseSafe++ | 76.1% | #9 of 15 | Official leaderboard | |
| nocot-bench (NCRI) | 49.4 | #168 of 185 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 63.3% | #46 of 46 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 67.3 | #63 of 152 | Official leaderboard | |
| Conceptual Reasoning Index | 49.0 | #73 of 151 | Official leaderboard | |
| LMCA | 36.9% | #74 of 147 | Official leaderboard | |
| CAIS Risk Index | 56.8 | #45 of 67 | Official leaderboard | |
| SAGE (Vals) | 38.1% | #61 of 81 | Official leaderboard | |
| Vals Multimodal Index | 47.6% | #28 of 33 | Official leaderboard | |
| CAIS Vision Capabilities Index | 38.9 | #45 of 53 | Official leaderboard | |
| CAIS Text Capabilities Index | 16.1 | #58 of 68 | Official leaderboard |
The Self-Reported Index fits one score to the 11 capability results OpenAI 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 139.6 to 146.0. How we compute it.
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 |
|---|---|---|---|
| OpenAI MRCR v2 | 44.2% OpenAI MRCR v2 8-needle 64K-128K | 33% Leaderboard, 8-needle, 128K to 256K bin, xhigh reasoning | +11 pts |
| GPQA Diamond | 82.8% GPQA Diamond | 78.5% Epoch AI run, high | +4.3 pts |
| Toolathlon | 35.5% Toolathlon | 35.5% Leaderboard, xhigh | Same |
OpenAI'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)
The benchmarks the lab chose to headline, with the scores it gave.