Gemini 3.8 Flash ranks first on 2 of the 325 benchmarks with five or more models scored.
Google DeepMind Released September 2nd, 2026 Family Gemini 3.8 The lab's model card Compare with others Google DeepMind Sep 2026 Closed weights Mid-size
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 | |
|---|---|---|---|---|
| LMArena (Elo) | 1,497 | #9 of 335 | Official leaderboard | |
| SEA-HELM | 77.3 | #2 of 71 | Official leaderboard | |
| Extended NYT Connections | 97.4% | #6 of 103 | Official leaderboard | |
| SimpleQA Verified | 74.6% | #4 of 56 | Official leaderboard | |
| Antidote | 1,087 | #3 of 36 | Official leaderboard | |
| FACTS Multimodal | 57.6% | #2 of 23 | Official leaderboard | |
| Hemingway-bench | 1,093 | #4 of 42 | Official leaderboard | |
| Humanity's Last Exam | 44.5% | #5 of 47 | Official leaderboard | |
| Last Translation Benchmark | 39.9 | #4 of 36 | Official leaderboard | |
| Global MMLU Lite | 95.3% | #5 of 43 | Official leaderboard | |
| DesignArena ASCII Art | 1,273 | #11 of 77 | Official leaderboard | |
| FACTS Parametric | 77.3% | #7 of 45 | Official leaderboard | |
| Creative Writing v3 | 78.0 | #22 of 120 | Official leaderboard | |
| StudyArena Text | 1,265 | #17 of 85 | Official leaderboard | |
| Joke Arena | 1,546 | #5 of 25 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 76.8 | #25 of 119 | Official leaderboard | |
| FACTS Parametric V2 | 78.7% | #6 of 27 | Official leaderboard | |
| FACTS Benchmark Suite | 71.2% | #7 of 23 | Official leaderboard | |
| ELI5 Arena | 1,583 | #6 of 18 | Official leaderboard | |
| Document Arena | 1,469 | #15 of 39 | Official leaderboard | |
| Corp Speak Arena | 1,542 | #9 of 22 | Official leaderboard | |
| HLE Extended | 36.8% | #13 of 31 | Official leaderboard | |
| ComplexConstraints | 63.2% | #7 of 15 | Official leaderboard | |
| FACTS Grounding v2 | 68.2% | #12 of 25 | Official leaderboard | |
| Lech Mazur Creative Writing | 0.08 | #25 of 50 | Official leaderboard | |
| HLE-Verified | 54.9% | #2 of 4 | Lab model card | |
| AA-LCR Extended | 55.6% | #33 of 49 | Official leaderboard | |
| HLE-Diamond | 33.3% | #15 of 20 | Official leaderboard | |
| HANDBOOK.md | 26.5% | #12 of 16 | Official leaderboard | |
| FACTS Search V2 | 76.3% | #17 of 21 | Official leaderboard | |
| Opening Lines Arena | 1,392 | #14 of 15 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ARC-AGI | 98.5% | #2 of 95 | Official leaderboard | |
| GPQA Diamond | 95.4% | #5 of 229 | Epoch AI run | |
| SimpleBench | 82.4% | #9 of 106 | Official leaderboard | |
| DTBench | 95.7% | #21 of 180 | Official leaderboard | |
| ARC-AGI-2 | 89.2% | #12 of 98 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 98.9% | #26 of 212 | Epoch AI run | |
| Surface Evolver Bench | 76.9% | #7 of 25 | Official leaderboard | |
| MysteryMechanism | 36.5% | #9 of 31 | Official leaderboard | |
| Prophet Arena | 86.0 | #7 of 24 | Official leaderboard | |
| xbench-ScienceQA | 72.4% | #10 of 29 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 68.4% | #31 of 86 | Official leaderboard | |
| ExplorationBench | 79% | #4 of 10 | Official leaderboard | |
| Leipzig Benchmark | 41% | #10 of 19 | Official leaderboard | |
| Terminal-Bench-Science | 12.4% | #9 of 17 | Official leaderboard | |
| Human Behavior Bench | 79.2 | #6 of 11 | Official leaderboard | |
| Riemann-bench | 56.8% | #12 of 20 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 79.5% | #14 of 23 | Official leaderboard | |
| ProofBench v1.1 | 48% | #33 of 52 | Official leaderboard | |
| LABBench2 | 86.2% | #2 of 3 | Lab model card | |
| FrontierMath Tier 4 v2 | 21.9% | #47 of 68 | Official leaderboard | |
| LatchBio Capabilities v1.0 | 37.7% | #19 of 27 | Official leaderboard | |
| BioMysteryBench | 62.2% | #25 of 30 | Official leaderboard | |
| SciUniverse | 14.6% | #5 of 6 | Official leaderboard | |
| FrontierMath Erdős | 0% | #14 of 15 | Official leaderboard | |
| FrontierPhysics | 0% | #21 of 22 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 34% | #25 of 26 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WeirdML | 84.8% | #8 of 112 | Official leaderboard | |
| DeepSWE | 73.8% | #2 of 28 | Official leaderboard | |
| StructEval | 80.6% | #2 of 28 | Official leaderboard | |
| ALE-Bench | 2,182 | #14 of 126 | Official leaderboard | |
| DesignArena SVG | 1,291 | #15 of 103 | Official leaderboard | |
| DesignArena UI Component | 1,306 | #21 of 139 | Official leaderboard | |
| DesignArena Website | 1,292 | #24 of 151 | Official leaderboard | |
| SWE Atlas Test Writing | 53.7% | #4 of 22 | Official leaderboard | |
| DesignArena Code | 1,293 | #27 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,297 | #28 of 143 | Official leaderboard | |
| DesignArena Data Visualization | 1,277 | #25 of 118 | Official leaderboard | |
| DesignArena 3D Design | 1,278 | #30 of 136 | Official leaderboard | |
| SWE-Together | 64.2% | #5 of 22 | Official leaderboard | |
| DesignArena React Native | 1,228 | #13 of 56 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,251 | #6 of 24 | Official leaderboard | |
| DAXBench v2 | 85.6% | #12 of 47 | Official leaderboard | |
| WebDev Arena | 1,568 | #27 of 105 | arena.ai, via Epoch AI | |
| SWE Atlas | 47% | #6 of 22 | Official leaderboard | |
| DesignArena Image to HTML | 1,230 | #14 of 51 | Official leaderboard | |
| Code Arena | 1,583 | #37 of 128 | Official leaderboard | |
| Vibe Code Bench | 78.7% | #36 of 120 | Official leaderboard | |
| DesignArena Full Stack | 1,220 | #15 of 49 | Official leaderboard | |
| Image-to-WebDev Arena | 1,594 | #17 of 55 | Official leaderboard | |
| DesignArena Video to Website | 1,211 | #6 of 19 | Official leaderboard | |
| ReactBench | 34.4% | #9 of 28 | Official leaderboard | |
| DesignArena Agentic Overall | 1,217 | #24 of 74 | Official leaderboard | |
| DesignArena Frontend Web App | 1,222 | #19 of 53 | Official leaderboard | |
| SWE Atlas Refactoring | 44.8% | #7 of 17 | Official leaderboard | |
| IOI v2 (Vals) | 56.9% | #20 of 45 | Official leaderboard | |
| Code Migration | 36.5% | #36 of 78 | Official leaderboard | |
| SWE-bench CLI | 6.30% | #8 of 16 | Official leaderboard | |
| Next.js Agent Evals | 93% | #7 of 14 | Official leaderboard | |
| DesignArena Blender | 1,192 | #6 of 12 | Official leaderboard | |
| Vibe Code Bench 1-100 | 18.8% | #14 of 26 | Official leaderboard | |
| SWE-bench Verified Extended | 54% | #24 of 44 | Official leaderboard | |
| Floorplan Arena | 1,482 | #6 of 11 | Official leaderboard | |
| ScarfBench | 66% | #6 of 11 | Official leaderboard | |
| Snorkel Agentic Coding 2.0 | 32.6% | #9 of 16 | Official leaderboard | |
| FDE-Gym | 0.19 | #7 of 12 | Official leaderboard | |
| Chart Arena | 1,472 | #9 of 15 | Official leaderboard | |
| MirrorCode | 38.9% | #24 of 39 | Epoch AI run | |
| APEX-SWE | 36.4% | #42 of 65 | Official leaderboard | |
| SWE Atlas v1.1 | 57.3% | #19 of 28 | Official leaderboard | |
| RSI Arena | 981 | #7 of 10 | Official leaderboard | |
| Vals RSI Index | 0.20 | #19 of 27 | Official leaderboard | |
| WeirdML v3 | 7.38% | #20 of 27 | Official leaderboard | |
| AutoResearchExam | 0.48 | #9 of 12 | Official leaderboard | |
| FrontierSWE | 19.6% | #19 of 25 | Official leaderboard | |
| Code Migration v1.1 | 38.6% | #23 of 30 | Official leaderboard | |
| RSI-Exam | 34.1% | #14 of 18 | Official leaderboard | |
| SWE-Bench Pro V2 | 94.9% | #14 of 18 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,086 | #37 of 46 | Official leaderboard | |
| CursorBench | 39.6% | #17 of 21 | cursor.com, via Epoch AI | |
| SWE-Bench Pro V2 Hard | 58.8% | #15 of 18 | Official leaderboard | |
| MacroscopeBench | 70.7% | #14 of 16 | Official leaderboard | |
| ViBench 2.0 | 0% | #8 of 9 | Official leaderboard | |
| OpenDesign Arena | 68.9 | #19 of 21 | Official leaderboard | |
| Android Bench 2.0 | 8.00% | #16 of 17 | Official leaderboard | |
| sudo L7 | 10.6% | #17 of 18 | Official leaderboard | |
| Landing Page Arena | 1,340 | #14 of 14 | Official leaderboard | |
| Web Clone Bench | 1.00% | #10 of 10 | Official leaderboard | |
| DeepRSI | 643 | #8 of 8 | Official leaderboard | |
| FrontierCS 1.1 | 47.8 | #6 of 6 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Game Arena Dark Hex | 581 | #1 of 41 | Official leaderboard | |
| Game Arena Chess | 1,458 | #2 of 42 | Official leaderboard | |
| LLM Chess | 1,546 | #6 of 123 | Official leaderboard | |
| GameHorizon | 77.3 | #2 of 37 | Official leaderboard | |
| Game Arena Reversi | 935 | #3 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 | 89.4% | #5 of 56 | Lab model card | |
| GameWorld | 63.5% | #2 of 22 | Official leaderboard | |
| Game Arena Checkers | 1,238 | #4 of 41 | Official leaderboard | |
| Game Arena Coin Game | 484 | #4 of 39 | Official leaderboard | |
| Game Arena Four in a Row | 698 | #5 of 43 | Official leaderboard | |
| Chess Puzzles | 58% | #14 of 93 | Epoch AI run | |
| Kaggle Game Arena | 432 | #7 of 42 | Official leaderboard | |
| RuneBench | 5.96 | #13 of 76 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 44.4% | #8 of 44 | Official leaderboard | |
| Mystery Game Puzzles | 47% | #17 of 84 | Epoch AI run | |
| BabaIsBench | 100% | #7 of 32 | Official leaderboard | |
| Game Arena Dots and Boxes | 1,018 | #9 of 40 | Official leaderboard | |
| Game Arena Word Association | 344 | #9 of 40 | Official leaderboard | |
| Game Arena Word Art | 1,389 | #9 of 39 | Official leaderboard | |
| AutomationBench | 29.7% | #10 of 42 | Official leaderboard | |
| AdaExploreBench | 67.3% | #2 of 8 | Official leaderboard | |
| Realtime-Bench | 27.3% | #2 of 8 | Official leaderboard | |
| Game Arena Clobber | 477 | #10 of 39 | Official leaderboard | |
| ARC-AGI-3 | 35% | #7 of 27 | Official leaderboard | |
| Game Arena Go | 1,170 | #11 of 39 | Official leaderboard | |
| Game Arena Nine Men's Morris | 788 | #11 of 39 | Official leaderboard | |
| BrowseComp Extended | 41.5% | #12 of 42 | Official leaderboard | |
| Cresta-Bench | 76.1% | #4 of 14 | Official leaderboard | |
| ApprenticeBench | 24% | #6 of 18 | Official leaderboard | |
| Game Arena Poker | 4.68 | #11 of 31 | Official leaderboard | |
| Agent Arena | 2.30% | #20 of 56 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 380 | #14 of 39 | Official leaderboard | |
| Game Arena Bargaining | 208 | #16 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 431 | #17 of 39 | Official leaderboard | |
| DuelLab GameBench 2 | 64.0 | #28 of 57 | Official leaderboard | |
| Time Horizon Index: KSP | 18.8% | #7 of 13 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 51.5% | #10 of 18 | Official leaderboard | |
| OSWorld 2.0 | 59% | #11 of 19 | Lab model card | |
| WitnessBench | 6.70% | #7 of 12 | Official leaderboard | |
| Long-Horizon-Terminal-Bench 2.0 | 47.4% | #9 of 15 | Official leaderboard | |
| Terminal-Bench 4.0 | 19.1% | #26 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 18.2% | #28 of 43 | Official leaderboard | |
| Hermes Bench | 62.2% | #15 of 17 | Official leaderboard | |
| CUA-bench | 4.17% | #9 of 9 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vals Finance Agent v2 | 61.4% | #2 of 80 | Official leaderboard | |
| LegalBench | 87% | #8 of 138 | Official leaderboard | |
| Excel Modeling Benchmark | 72.2% | #8 of 76 | Official leaderboard | |
| FailureSensorIQ | 69% | #3 of 24 | Official leaderboard | |
| Vals Index v2 | 62.3% | #10 of 65 | Official leaderboard | |
| Vals Finance Agent v2.1 | 65.8% | #8 of 40 | Official leaderboard | |
| APEX-1 | 68.8% | #15 of 71 | Official leaderboard | |
| Legora BAR | 103.0 | #4 of 17 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 10% | #19 of 80 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 72.9% | #10 of 42 | Official leaderboard | |
| APEX-Agents | 64.3% | #15 of 61 | Collected by Epoch AI, source not named | |
| EnterpriseBench: CoreCraft | 58.5% | #13 of 51 | Official leaderboard | |
| OR-License | 82.6% | #3 of 11 | Official leaderboard | |
| TaxEval v2 | 74.5% | #36 of 129 | Official leaderboard | |
| Tax Agent Bench | 32.1% | #25 of 84 | Official leaderboard | |
| RCM-Bench-1 | 90.2% | #6 of 20 | Official leaderboard | |
| MedCode | 48.1% | #30 of 96 | Official leaderboard | |
| MedXpertQA MM Extended | 60.8% | #13 of 41 | Official leaderboard | |
| APEX-Agents 1.1 | 72.9% | #21 of 65 | Official leaderboard | |
| MMEnterprise-Bench | 63.9% | #3 of 9 | Official leaderboard | |
| Vals Finance Index | 60.5% | #14 of 40 | Official leaderboard | |
| MedScribe | 84.5% | #34 of 97 | Official leaderboard | |
| FrontierFinance | 50% | #6 of 17 | Official leaderboard | |
| PRBench | 48.9% | #17 of 48 | Official leaderboard | |
| Vals Index v2.1 | 54.9% | #18 of 49 | Official leaderboard | |
| Legal Research Bench | 38.9% | #39 of 93 | Official leaderboard | |
| PolicyBench | 85.6% | #21 of 50 | Official leaderboard | |
| Public Benefits Bench | 65.3% | #22 of 50 | Official leaderboard | |
| Physical-Engineering's Last Exam | 0.28 | #8 of 18 | Official leaderboard | |
| Parametric CAD Bench | 36.2% | #5 of 11 | Official leaderboard | |
| DesignArena PDF | 1,199 | #17 of 37 | Official leaderboard | |
| MortgageTax | 65.3% | #41 of 87 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 12.1% | #19 of 38 | Official leaderboard | |
| TaxCalcBench v2 | 46.3% | #16 of 32 | Official leaderboard | |
| EEBench | 55.4% | #15 of 30 | Official leaderboard | |
| CAD Arena | 0.41 | #6 of 12 | Official leaderboard | |
| Hebbia Financial Services Benchmark | 60.9% | #5 of 10 | Official leaderboard | |
| FrontierCAD Assemblies | 0.37 | #3 of 6 | Official leaderboard | |
| APEX-Accounting | 58.8% | #25 of 49 | Official leaderboard | |
| GDPval-AA v2 | 1,545 | #14 of 27 | Lab model card | |
| Vending-Bench 2 | $5,094 | #35 of 67 | Official leaderboard | |
| Vals Legal Index | 69.8% | #20 of 38 | Official leaderboard | |
| Clipping Bench | $596 | #10 of 19 | Official leaderboard | |
| Remote Labor Index | 5.83% | #10 of 19 | Official leaderboard | |
| Quant-Arena | 43.0 | #5 of 9 | Official leaderboard | |
| GDP.xlsx | 18.3% | #12 of 20 | Official leaderboard | |
| Build-A-Chip-Bench | 0.80% | #3 of 5 | Official leaderboard | |
| Vals Index v2.2 | 56% | #23 of 37 | Official leaderboard | |
| Digital Work Index | 16.9 | #10 of 16 | Official leaderboard | |
| ATLAS-Finance | 3.67% | #7 of 11 | Official leaderboard | |
| ATLAS-Finance v1.1.0 | 3.67% | #7 of 11 | Official leaderboard | |
| DAYJOB: Healthcare | 2.00% | #24 of 37 | Official leaderboard | |
| DesignArena Spreadsheet | 1,094 | #23 of 35 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,132 | #24 of 36 | Official leaderboard | |
| FrontierCAD Parts | 0.33 | #4 of 6 | Official leaderboard | |
| GDPval Extended | 0.60% | #29 of 43 | Official leaderboard | |
| DAYJOB: Finance | 4.50% | #27 of 39 | Official leaderboard | |
| LegalReasonBench | 39.6% | #7 of 10 | Official leaderboard | |
| Tuesday Work Index | 43.2 | #31 of 44 | Official leaderboard | |
| DAYJOB: Tax | 0.80% | #11 of 15 | Official leaderboard | |
| Ivo-micro1 ContractBench | 55% | #6 of 8 | Official leaderboard | |
| SheetBench | 45.6% | #6 of 8 | Official leaderboard | |
| Bedside Bench | 86.5% | #13 of 17 | Official leaderboard | |
| Immigration Bench | 63.4% | #20 of 25 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 36.5% | #12 of 15 | Official leaderboard | |
| WorkplaceAgents | 11.2% | #12 of 14 | Official leaderboard | |
| SnorkelLegal | 25.3% | #14 of 16 | Official leaderboard | |
| SnorkelUnderwrite 2.0 | 19.1% | #14 of 16 | Official leaderboard | |
| DAYJOB: Accounting | 6.00% | #15 of 17 | Official leaderboard | |
| Genspark Slides Benchmark | 53.4% | #9 of 10 | Official leaderboard | |
| BlenderWorld | 34.2 | #12 of 13 | Official leaderboard | |
| SnorkelFinance 2.0 | 5.40% | #15 of 16 | Official leaderboard | |
| SnorkelManufacturing | 7.10% | #15 of 16 | Official leaderboard | |
| SnorkelRevOps | 5.40% | #16 of 16 | Official leaderboard | |
| PharmaBench (micro1) | 5.60% | #13 of 13 | Official leaderboard | |
| RealmPathology | 66.3% | #13 of 13 | Official leaderboard | |
| MissionReport | 54.3% | #10 of 10 | Official leaderboard | |
| Short Form Bench | $1,940 | #9 of 9 | Official leaderboard | |
| micro1 Medical Index | 60% | #8 of 8 | Official leaderboard | |
| Epoch Automation Reports | 42.3% | #6 of 6 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Audio MC | 60.4% | #1 of 18 | Official leaderboard | |
| LVBench | 87.8% | #2 of 31 | Lab model card | |
| Vision Arena | 1,293 | #9 of 136 | Official leaderboard | |
| GeoBench | 86% | #4 of 48 | Official leaderboard | |
| 1H-VideoQA | 87.1% | #2 of 20 | Official leaderboard | |
| ArtifactArena | 1,223 | #3 of 21 | Official leaderboard | |
| ParseBench | 71.2% | #7 of 44 | Official leaderboard | |
| ClockBench | 66.7% | #6 of 37 | Official leaderboard | |
| Chartography | 40.9% | #7 of 43 | Official leaderboard | |
| CharXiv | 86.2% | #17 of 75 | Lab model card | |
| Humanity's Sixth Sense | 41.6% | #8 of 34 | Official leaderboard | |
| vCX Benchmark | 90.6% | #10 of 42 | Official leaderboard | |
| MineBench | 1,789 | #19 of 75 | Official leaderboard | |
| Blueprint-Bench 2 | 38.6% | #11 of 38 | Official leaderboard | |
| Code2Video-Compass | 5.53 | #5 of 15 | Official leaderboard | |
| OmniParseBench | 92.9 | #3 of 9 | Official leaderboard | |
| VoxMem | 34% | #3 of 9 | Official leaderboard | |
| MMMU-Pro Extended | 36.4% | #15 of 38 | Official leaderboard | |
| WorldAuditBench | 32.4% | #2 of 5 | Official leaderboard | |
| VoxelBench | 1,810 | #23 of 56 | Official leaderboard | |
| DesignArena Code to Video | 1,189 | #12 of 29 | Official leaderboard | |
| 4DCodeBench | 0.64 | #7 of 16 | Official leaderboard | |
| Dual-Arm Sim | 38.8 | #5 of 11 | Official leaderboard | |
| BVB | 65.9 | #25 of 45 | Official leaderboard | |
| RoboDojo-RC Tier 1 | 36.2% | #5 of 9 | Official leaderboard | |
| CharXiv Extended | 78.9% | #22 of 38 | Official leaderboard | |
| GDP.pdf | 44.2% | #11 of 19 | Official leaderboard | |
| CineRecon | 47.1 | #7 of 12 | Official leaderboard | |
| Embodied First Exam | 6.36% | #6 of 9 | Official leaderboard | |
| Furniture Assembly Benchmark | 31.7% | #26 of 35 | Official leaderboard | |
| RobotWorld | 1.19% | #4 of 5 | Official leaderboard | |
| CodeCap-Bench | 47.7 | #6 of 7 | Official leaderboard | |
| VibeVideoBench | 2.80 | #8 of 9 | Official leaderboard | |
| Code2Video Arena | 1,413 | #12 of 13 | Official leaderboard | |
| Code2Video Benchmark | 1,440 | #18 of 19 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 125.8 | #7 of 185 | Official leaderboard | |
| World-Class Bio | 57.1% | #13 of 74 | Official leaderboard | |
| BioSecBench-Refusal | 47.4% | #5 of 22 | Official leaderboard | |
| Human Pathogen Capabilities Test | 62.4% | #20 of 86 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 64.4% | #18 of 71 | Official leaderboard | |
| BioTIER Refuse | 78% | #22 of 69 | Official leaderboard | |
| SRE-Bench | 11.1% | #12 of 35 | Official leaderboard | |
| TxBench-Antibody-Discovery | 49% | #12 of 30 | Official leaderboard | |
| scBench-Long | 19.9% | #12 of 27 | Official leaderboard | |
| VIALS 2 | 29.4% | #7 of 14 | Official leaderboard | |
| BioSimBench | 0.15 | #6 of 12 | Official leaderboard | |
| Suicide Memory Bench | 0.80 | #6 of 12 | Official leaderboard | |
| Molecular Biology Capabilities Test | 48.9% | #46 of 90 | Official leaderboard | |
| BioSecBench-Function | 36% | #14 of 27 | Official leaderboard | |
| SpatialBench-Long | 28.6% | #14 of 27 | Official leaderboard | |
| VCT-v2 | 45.2% | #13 of 25 | Official leaderboard | |
| BioTIER Permit | 99.4% | #36 of 69 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 46.3% | #26 of 46 | Official leaderboard | |
| TxBench-Oligo-Discovery | 35.1% | #17 of 30 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 54.7% | #17 of 30 | Official leaderboard | |
| DistressBench | 70.7% | #24 of 42 | Official leaderboard | |
| BioSecBench-Surveillance | 44.1% | #16 of 27 | Official leaderboard | |
| EpiBench | 21.4% | #18 of 27 | Official leaderboard | |
| VariantBench | 31.4% | #18 of 27 | Official leaderboard | |
| SAFE-Teen Relationship boundaries | 63% | #13 of 17 | Official leaderboard | |
| OncologyBench | 13.3% | #23 of 30 | Official leaderboard | |
| SecIT Bench | 69% | #20 of 26 | Official leaderboard | |
| dfbench | 44.2% | #15 of 18 | Official leaderboard | |
| AI Wellbeing Index percent positive experiences | 68.5% | #5 of 6 | Official leaderboard | |
| MetagenomicsBench | 26.7% | #22 of 26 | Official leaderboard | |
| SAFE-Teen Health advice | 69% | #15 of 17 | Official leaderboard | |
| CyberBench (Vals) | 43.8% | #23 of 26 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 43.8% | #40 of 44 | Official leaderboard | |
| WellnessBench | 47% | #11 of 12 | Official leaderboard | |
| SAFE-Teen Urgent safety | 56% | #17 of 17 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CAIS Vision Capabilities Index | 69.9 | #6 of 53 | Official leaderboard | |
| ACCoRD | 78.2 | #19 of 152 | Official leaderboard | |
| CAIS Text Capabilities Index | 50.2 | #10 of 68 | Official leaderboard | |
| LMCA | 52.9% | #23 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 65.0 | #24 of 151 | Official leaderboard | |
| Political Science LLM Benchmark | 68.7% | #7 of 42 | Official leaderboard | |
| SAGE v1.1 (Vals) | 88.2% | #9 of 29 | Official leaderboard | |
| CAIS Risk Index | 49.3 | #31 of 67 | Official leaderboard | |
| Hermes Index | 38.1 | #9 of 17 | Official leaderboard | |
| Vals Science Index | 35.8% | #15 of 25 | Official leaderboard | |
| Xbench | 22.9 | #13 of 17 | Official leaderboard | |
| LiveBench | 57.3% | #24 of 31 | Official leaderboard | |
| SAGE (Vals) | 35.1% | #66 of 81 | Official leaderboard |
The Self-Reported Index fits one score to the 10 capability results Google DeepMind 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 155.5 to 163.9. 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 |
|---|---|---|---|
| BioMysteryBench | 88.8% Best of BioMysteryBench, Human Solvable / BioMysteryBench, Human Difficult | 62.2% Leaderboard, Vals AI harness, v11 full set | +27 pts |
| GDP.pdf | 35% GDP.PDF (All pass rate) | 44.2% Leaderboard, high | −9.2 pts |
| DeepSWE | 73.7% DeepSWE v1.1 | 73.8% Leaderboard, high, mini-swe-agent | Same |
| Vals Finance Agent v2 | 61.4% Vals Finance Agent v2 | 61.4% Leaderboard | Same |
| Terminal-Bench 4.0 | 19.1% Terminal-bench 4.0 | 19.1% Leaderboard, high, mini-SWE-agent | Same |
| Harvey LAB (Legal Agent Benchmark) | 10% Harvey's Legal Agent Benchmark (All pass rate) | 10% Leaderboard | Same |
Google DeepMind'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.
Every result in the longer documents the lab published with it. 5 of them aren't in the launch post.