Gemini 3.5 Flash ranks first on 2 of the 225 benchmarks with five or more models scored.
Google DeepMind May 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 | |
|---|---|---|---|---|
| Wer wird Millionär? LLM Benchmark | 433333.0 | #3 of 77 | Official leaderboard | |
| HUMAINE | 35 | #2 of 50 | Official leaderboard | |
| Global MMLU Lite | 95.4% | #2 of 43 | Official leaderboard | |
| KINA | 52.1% | #2 of 37 | Official leaderboard | |
| LMArena (Elo) | 1,478 | #28 of 335 | Official leaderboard | |
| SEA-HELM | 76.3 | #7 of 71 | Official leaderboard | |
| Antidote | 1,079 | #5 of 36 | Official leaderboard | |
| SimpleQA Verified | 70.4% | #8 of 56 | Official leaderboard | |
| Hemingway-bench | 1,071 | #7 of 42 | Official leaderboard | |
| Extended NYT Connections | 92.6% | #18 of 103 | Official leaderboard | |
| FACTS Parametric | 73.5% | #9 of 45 | Official leaderboard | |
| DesignArena ASCII Art | 1,248 | #21 of 77 | Official leaderboard | |
| OpenAI MRCR v2 | 54.6% | #14 of 47 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 71.8 | #36 of 119 | Official leaderboard | |
| MRCR v2 | 77.3% | #11 of 30 | Lab model card | |
| Judgemark v4 | 68.8 | #16 of 40 | Official leaderboard | |
| Humanity's Last Exam | 40.2% | #65 of 154 | Lab model card | |
| Document Arena | 1,459 | #24 of 39 | Official leaderboard | |
| CompassArena | 994 | #7 of 10 | Official leaderboard | |
| HLE Extended | 14.8% | #22 of 31 | Official leaderboard | |
| BullshitBench V2 | 20% | #95 of 129 | Official leaderboard | |
| FACTS Search | 62.6% | #24 of 31 | Official leaderboard | |
| Lech Mazur Creative Writing | -2.03 | #40 of 50 | Official leaderboard | |
| AA-LCR Extended | 39.9% | #41 of 49 | Official leaderboard | |
| EQ-Bench 4 | 1,077 | #28 of 32 | Official leaderboard | |
| FACTS Search V2 | 64.3% | #19 of 21 | Official leaderboard | |
| BullshitBench | 9.09% | #100 of 107 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveMathematicianBench | 45.7% | #1 of 17 | Official leaderboard | |
| GPQA Diamond | 92.8% | #26 of 229 | Epoch AI run | |
| SimpleBench | 76.7% | #15 of 106 | Official leaderboard | |
| DTBench | 94.7% | #26 of 180 | Official leaderboard | |
| FrontierMath | 39% | #9 of 55 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 32.3% | #9 of 47 | Official leaderboard | |
| HMMT 2026 | 95.5% | #6 of 31 | Official leaderboard | |
| MathArena Apex Shortlist | 82.5% | #8 of 38 | Official leaderboard | |
| SpatialBench (LatchBio) | 55.6% | #3 of 14 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 95.6% | #47 of 212 | Epoch AI run | |
| ARC-AGI-2 | 72.1% | #26 of 98 | Official leaderboard | |
| FrontierMath Tier 4 | 14.6% | #16 of 55 | Epoch AI run | |
| ARC-AGI | 92.5% | #28 of 95 | Official leaderboard | |
| NatureBench | 12.2% | #7 of 18 | Official leaderboard | |
| ProofBench | 29% | #24 of 59 | Official leaderboard | |
| ForecastBench | 0.13 | #35 of 82 | Official leaderboard | |
| Surface Evolver Bench | 58.1% | #11 of 25 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 62.8% | #40 of 86 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 26.8% | #39 of 68 | Official leaderboard | |
| AIME 2026 | 95% | #18 of 31 | Official leaderboard | |
| BixBench3 | 42.1 | #12 of 20 | Official leaderboard | |
| ResearchClawBench | 17.9 | #15 of 23 | Official leaderboard | |
| ProofBench v1.1 | 31% | #36 of 52 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 68.5% | #17 of 23 | Official leaderboard | |
| Project Euler (MathArena) | 82% | #3 of 4 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 79.3% | #3 of 32 | Epoch AI run | |
| OpenHands Index | 62.6% | #7 of 33 | Official leaderboard | |
| DesignArena SVG | 1,245 | #23 of 103 | Official leaderboard | |
| WeirdML | 62.6% | #28 of 112 | Official leaderboard | |
| DesignArena Website | 1,261 | #45 of 151 | Official leaderboard | |
| DesignArena Game Dev | 1,254 | #43 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,266 | #42 of 139 | Official leaderboard | |
| DesignArena Code | 1,257 | #47 of 143 | Official leaderboard | |
| ALE-Bench | 1,712 | #43 of 126 | Official leaderboard | |
| DesignArena 3D Design | 1,227 | #48 of 136 | Official leaderboard | |
| DesignArena Data Visualization | 1,238 | #45 of 118 | Official leaderboard | |
| WebDev Arena | 1,506 | #44 of 105 | arena.ai, via Epoch AI | |
| AndroidBench | 71.1% | #15 of 32 | Official leaderboard | |
| DesignArena Full Stack | 1,175 | #23 of 49 | Official leaderboard | |
| Code Arena | 1,500 | #61 of 128 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,174 | #22 of 46 | Official leaderboard | |
| DesignArena Agentic Overall | 1,166 | #38 of 74 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,129 | #14 of 27 | Official leaderboard | |
| DesignArena Video to Website | 1,181 | #10 of 19 | Official leaderboard | |
| Vibe Code Bench | 48.7% | #66 of 120 | Official leaderboard | |
| DesignArena Frontend Web App | 1,182 | #30 of 53 | Official leaderboard | |
| DesignArena React Native | 1,170 | #32 of 56 | Official leaderboard | |
| Code Migration | 26.8% | #45 of 78 | Official leaderboard | |
| SWE-Marathon | 7.00% | #7 of 12 | Official leaderboard | |
| DesignArena Image to HTML | 1,189 | #32 of 51 | Official leaderboard | |
| SWE-Bench Pro | 55.1% | #51 of 81 | Lab model card | |
| ProgramBench tests passed | 53.6% | #16 of 25 | Official leaderboard | |
| APEX-SWE | 36.1% | #43 of 65 | Official leaderboard | |
| GBAEval | 6.70% | #14 of 21 | Official leaderboard | |
| Image-to-WebDev Arena | 1,506 | #37 of 55 | Official leaderboard | |
| ProgramBench almost solved | 3.00% | #17 of 25 | Official leaderboard | |
| DesignArena Image to Frontend | 1,139 | #23 of 31 | Official leaderboard | |
| ProgramBench | 0% | #20 of 25 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 2.70% | #31 of 38 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,081 | #21 of 24 | Official leaderboard | |
| SWE-bench Verified Extended | 22.7% | #39 of 44 | Official leaderboard | |
| DeepSWE | 36.1% | #25 of 28 | Official leaderboard | |
| ReactBench | 23.1% | #25 of 28 | Official leaderboard | |
| Monthly-SWEBench | 29% | #11 of 12 | Official leaderboard | |
| Senior SWE-Bench | 6.32% | #22 of 23 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Toolathlon | 56.5% | #1 of 50 | Official leaderboard | |
| SnakeBench | 34 | #13 of 188 | Official leaderboard | |
| Pencil Puzzle Bench | 41.9% | #6 of 53 | Official leaderboard | |
| Game Arena Dark Hex | 439 | #5 of 41 | Official leaderboard | |
| Clocktower Radio | 1,698 | #3 of 23 | Official leaderboard | |
| Game Arena Chess | 1,342 | #6 of 42 | Official leaderboard | |
| CAR-bench | 66.7% | #2 of 12 | Official leaderboard | |
| GameHorizon | 74.4 | #7 of 37 | Official leaderboard | |
| LisanBench | 4123.0 | #28 of 133 | Official leaderboard | |
| LLM Chess | 953 | #26 of 123 | Official leaderboard | |
| Game Arena Four in a Row | 559 | #10 of 43 | Official leaderboard | |
| MCP Atlas | 83.6% | #9 of 37 | Official leaderboard | |
| Game Arena Five in a Row | 786 | #6 of 20 | Official leaderboard | |
| Game Arena Reversi | 714 | #12 of 39 | Official leaderboard | |
| OSWorld-Verified | 78.4% | #13 of 41 | Lab model card | |
| Chess Puzzles | 45% | #30 of 93 | Epoch AI run | |
| AndroidWorld | 93.1% | #3 of 9 | Official leaderboard | |
| PlanBench-XL | 52.2% | #3 of 9 | Official leaderboard | |
| RuneBench | 5.35 | #26 of 76 | Official leaderboard | |
| CEO-Bench | $75,126 | #7 of 19 | Official leaderboard | |
| Mystery Game Puzzles | 32% | #31 of 84 | Epoch AI run | |
| Game Arena Word Association | 295 | #15 of 40 | Official leaderboard | |
| Toolathlon-Verified | 67.3% | #10 of 25 | Official leaderboard | |
| Game Arena Coin Game | 384 | #16 of 39 | Official leaderboard | |
| Game Arena Word Art | 643 | #17 of 39 | Official leaderboard | |
| BabaIsBench | 88% | #15 of 32 | Official leaderboard | |
| PinchBench 2.0 | 74.2% | #27 of 57 | Official leaderboard | |
| Kaggle Game Arena | 323 | #21 of 42 | Official leaderboard | |
| DuelLab GameBench 2 | 63.2 | #29 of 57 | Official leaderboard | |
| Game Arena Dots and Boxes | 897 | #21 of 40 | Official leaderboard | |
| Learn2Play Bench | 72.0 | #6 of 11 | Official leaderboard | |
| Game Arena Checkers | 925 | #23 of 41 | Official leaderboard | |
| TextQuests | 18.0 | #16 of 28 | Official leaderboard | |
| Terminal-Bench 2.1 | 76.2% | #33 of 56 | Lab model card | |
| Game Arena Ultimate Tic-Tac-Toe | 261 | #23 of 39 | Official leaderboard | |
| AutomationBench | 13.4% | #25 of 42 | Official leaderboard | |
| Game Arena Go | 756 | #24 of 39 | Official leaderboard | |
| SkillsBench | 48.2% | #14 of 22 | Official leaderboard | |
| Game Arena Clobber | 211 | #25 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 336 | #25 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 28.6% | #30 of 44 | Official leaderboard | |
| Game Arena Nine Men's Morris | 525 | #29 of 39 | Official leaderboard | |
| GAIA | 24.9% | #49 of 64 | Official leaderboard | |
| Vals Web Search Index | 41.4% | #4 of 5 | Official leaderboard | |
| Game Arena Bargaining | 142 | #33 of 39 | Official leaderboard | |
| CLASHarness | 13.5% | #12 of 14 | Official leaderboard | |
| EBR-bench | 4.76% | #29 of 31 | Epoch AI run | |
| Game Arena Poker | -43.25 | #30 of 31 | Official leaderboard | |
| BrowseComp Extended | 0.70% | #41 of 42 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedCode | 55.8% | #6 of 96 | Official leaderboard | |
| Vals Finance Agent v2 | 57.9% | #13 of 80 | Official leaderboard | |
| TRLawBench | 9.58 | #6 of 34 | Official leaderboard | |
| ProfBench | 56% | #14 of 72 | Official leaderboard | |
| MortgageTax | 68.1% | #18 of 87 | Official leaderboard | |
| GDPval-AA | 1,656 | #4 of 16 | Lab model card | |
| 3DCodeArena | 1,114 | #4 of 14 | Official leaderboard | |
| TaxEval v2 | 74.4% | #37 of 129 | Official leaderboard | |
| CorpFin v2 | 64.7% | #36 of 119 | Official leaderboard | |
| Excel Modeling Benchmark | 63.5% | #28 of 76 | Official leaderboard | |
| LegalBench | 83.6% | #52 of 138 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 50.8% | #22 of 51 | Official leaderboard | |
| MedXpertQA MM Extended | 57.5% | #18 of 41 | Official leaderboard | |
| Vending-Bench 2 | $5,396 | #32 of 67 | Official leaderboard | |
| Vals Index v2 | 53.1% | #32 of 65 | Official leaderboard | |
| VWE-Bench | 47.4% | #3 of 6 | Official leaderboard | |
| FoodTruck Bench | $22,311 | #32 of 59 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 2.50% | #46 of 80 | Official leaderboard | |
| Legal Research Bench | 30.8% | #55 of 93 | Official leaderboard | |
| APEX-1 | 62.2% | #43 of 71 | Official leaderboard | |
| YC-Bench | $987,017 | #37 of 59 | Official leaderboard | |
| Tax Agent Bench | 21.6% | #53 of 84 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,137 | #23 of 36 | Official leaderboard | |
| Public Benefits Bench | 59.5% | #33 of 50 | Official leaderboard | |
| MedScribe | 76.6% | #68 of 97 | Official leaderboard | |
| E-Commerce Bench | 1.90 | #14 of 19 | Official leaderboard | |
| PolicyBench | 79.8% | #37 of 50 | Official leaderboard | |
| JobBench | 31.5% | #12 of 16 | Official leaderboard | |
| RedlineBench | 45.1% | #3 of 4 | Official leaderboard | |
| Vals Index v2.1 | 44.8% | #39 of 49 | Official leaderboard | |
| DesignArena PDF | 1,134 | #30 of 37 | Official leaderboard | |
| DesignArena HTML Slides | 1,038 | #23 of 28 | Official leaderboard | |
| EEBench | 34.3% | #25 of 30 | Official leaderboard | |
| $OneMillion-Bench | 17% | #11 of 13 | Official leaderboard | |
| DesignArena Spreadsheet | 1,014 | #32 of 35 | Official leaderboard | |
| CommerceAgentBench | 43% | #12 of 13 | Official leaderboard | |
| APEX-Agents | 27.5% | #57 of 61 | Collected by Epoch AI, source not named | |
| TaxCalcBench v2 | 2.00% | #30 of 32 | Official leaderboard | |
| APEX-Accounting | 31.4% | #46 of 49 | Official leaderboard | |
| GDPval Extended | 0% | #41 of 43 | Official leaderboard | |
| APEX-Agents 1.1 | 44.3% | #62 of 65 | Official leaderboard | |
| Clipping Bench | $218 | #19 of 19 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU-Pro | 83.6% | #3 of 90 | Lab model card | |
| GeoBench | 85% | #6 of 48 | Official leaderboard | |
| ParseBench | 73.2% | #6 of 44 | Official leaderboard | |
| Vision Arena | 1,281 | #20 of 136 | Official leaderboard | |
| Math Kangaroo 2025 (MathArena) | 89.9% | #4 of 22 | Official leaderboard | |
| Chartography | 35.9% | #9 of 43 | Official leaderboard | |
| ZeroBench | 22% | #19 of 76 | Official leaderboard | |
| 1H-VideoQA | 80.2% | #5 of 20 | Official leaderboard | |
| Omni-InteractScope Explicit Request | 20.2% | #1 of 4 | Official leaderboard | |
| CharXiv | 84.2% | #21 of 75 | Lab model card | |
| Blueprint-Bench 2 | 33.6% | #14 of 38 | Official leaderboard | |
| MineBench | 1,636 | #29 of 75 | Official leaderboard | |
| MMAU-Pro | 68.5% | #4 of 10 | Official leaderboard | |
| ClockBench | 34.4% | #15 of 37 | Official leaderboard | |
| CharXiv Extended | 81.3% | #16 of 38 | Official leaderboard | |
| PerceptionBench (Moonshot) | 52% | #7 of 16 | Official leaderboard | |
| Humanity's Sixth Sense | 32.6% | #16 of 34 | Official leaderboard | |
| DesignArena Code to Video | 1,148 | #14 of 29 | Official leaderboard | |
| Omni-InteractScope Proactive | 41.1% | #2 of 4 | Official leaderboard | |
| VoxelBench | 1,646 | #30 of 56 | Official leaderboard | |
| MMMU-Pro Extended | 28.8% | #29 of 38 | Official leaderboard | |
| EnigmaEval | 25.4% | #4 of 5 | Official leaderboard | |
| vCX Benchmark | 81.7% | #34 of 42 | Official leaderboard | |
| VoxelBench Image | 1,000 | #13 of 13 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Permit | 99.9% | #6 of 69 | Official leaderboard | |
| nocot-bench (NCRI) | 101.5 | #21 of 185 | Official leaderboard | |
| World-Class Bio | 56.7% | #14 of 74 | Official leaderboard | |
| BioTIER Refuse | 81.7% | #14 of 69 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 64.7% | #15 of 71 | Official leaderboard | |
| Human Pathogen Capabilities Test | 62.7% | #19 of 86 | Official leaderboard | |
| VCT-v2 | 46.5% | #8 of 25 | Official leaderboard | |
| CyberBench (Vals) | 69.3% | #13 of 26 | Official leaderboard | |
| DrugDiscoveryBench | 50% | #10 of 20 | Official leaderboard | |
| GeneBench Pro | 8.14% | #8 of 12 | Official leaderboard | |
| Chrome Production Commit Scanning (Google) | 55% | #2 of 3 | Official leaderboard | |
| Molecular Biology Capabilities Test | 45% | #61 of 90 | Official leaderboard | |
| DistressBench | 66.9% | #33 of 42 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 53.4% | #38 of 46 | Official leaderboard | |
| Big Sleep Evaluation (Google) | 36% | #3 of 3 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 49.9% | #18 of 81 | Official leaderboard | |
| CAIS Vision Capabilities Index | 64.6 | #12 of 53 | Official leaderboard | |
| Conceptual Reasoning Index | 59.8 | #35 of 151 | Official leaderboard | |
| LMCA | 47.1% | #35 of 147 | Official leaderboard | |
| CAIS Text Capabilities Index | 39.1 | #22 of 68 | Official leaderboard | |
| StudentBench | 14.7 | #4 of 11 | Official leaderboard | |
| ACCoRD | 66.6 | #65 of 152 | Official leaderboard | |
| Vals Multimodal Index | 62.9% | #16 of 33 | Official leaderboard | |
| CAIS Risk Index | 61.1 | #54 of 67 | Official leaderboard |
The Self-Reported Index fits one score to the 11 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 154.3 to 159.2. 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 |
|---|---|---|---|
| Vals Finance Agent v2 | 57.9% Finance Agent v2 | 57.9% Leaderboard | Same |
| ARC-AGI-2 | 72.1% ARC-AGI-2 | 72.1% Leaderboard, high | Same |
| Blueprint-Bench 2 | 33.6% Blueprint-Bench 2 (Normalized score) | 33.6% Leaderboard | Same |
| MCP Atlas | 83.6% MCP Atlas | 83.6% Leaderboard, high | Same |
| Toolathlon | 56.5% Toolathlon | 56.5% 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.