Gemma 4 31B IT ranks first on 5 of the 122 benchmarks with five or more models scored.
Google DeepMind Apr 2026 Open weights Open model
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 | |
|---|---|---|---|---|
| IFEval | 98.9% | #1 of 135 | Lab model card | |
| RULER | 96.4% | #2 of 42 | Lab model card | |
| BOUQuET | 35.0 | #1 of 18 | Official leaderboard | |
| BBEH micro average | 74.4% | #1 of 15 | Lab model card | |
| IFStruct | 95.9% | #1 of 10 | Official leaderboard | |
| SEA-HELM | 76.2 | #9 of 71 | Official leaderboard | |
| HELM Arabic Enterprise | 73.8 | #3 of 23 | Official leaderboard | |
| HELM Arabic | 82.3 | #4 of 28 | Official leaderboard | |
| MMLU-Pro | 85.2% | #24 of 167 | Lab model card | |
| Global PIQA | 84.9% | #7 of 44 | Official leaderboard | |
| LMArena (Elo) | 1,452 | #61 of 335 | Official leaderboard | |
| AraGen | 72.7% | #10 of 42 | Official leaderboard | |
| AgentIF (2026-10 setting) | 64.8% | #6 of 24 | Official leaderboard | |
| CulturalRiddles | 39.6% | #5 of 20 | Official leaderboard | |
| EvoArena PersonaMem-Evo | 52.9% | #2 of 8 | Official leaderboard | |
| IFMTBench | 79.6% | #2 of 8 | Official leaderboard | |
| MMMLU | 88.4% | #19 of 71 | Lab model card | |
| IFBench | 76% | #22 of 80 | Lab model card | |
| QIMMA | 60.4% | #7 of 25 | Official leaderboard | |
| FACTS Grounding v2 | 75.8% | #8 of 25 | Official leaderboard | |
| EQ-Bench 3 | 1,464 | #25 of 78 | Official leaderboard | |
| Judgemark v4 | 72.3 | #13 of 40 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 7.40% | #37 of 104 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 26311.0 | #28 of 77 | Official leaderboard | |
| GraphWalks | 82.3% | #10 of 27 | Lab model card | |
| Global MMLU Lite | 90.7% | #20 of 43 | Official leaderboard | |
| FACTS Search | 68.9% | #15 of 31 | Official leaderboard | |
| Creative Writing v3 | 68.0 | #61 of 120 | Official leaderboard | |
| Extended NYT Connections | 70.6% | #57 of 103 | Official leaderboard | |
| MRCR v2 | 66.4% | #17 of 30 | Lab model card | |
| Humanity's Last Exam | 26.5% | #88 of 154 | Lab model card | |
| Longform Creative Writing (EQ-Bench) | 56.5 | #74 of 119 | Official leaderboard | |
| LOFT | 79.5% | #5 of 8 | Lab model card | |
| BullshitBench | 21.8% | #68 of 107 | Official leaderboard | |
| FACTS Multimodal | 42.3% | #15 of 23 | Official leaderboard | |
| BullshitBench V2 | 25% | #85 of 129 | Official leaderboard | |
| Document Arena | 1,452 | #26 of 39 | Official leaderboard | |
| Lech Mazur Creative Writing | -1.91 | #39 of 50 | Official leaderboard | |
| EQ-Bench 4 | 1,110 | #26 of 32 | Official leaderboard | |
| StudyArena Text | 1,188 | #70 of 85 | Official leaderboard | |
| SimpleQA Verified | 10.7% | #47 of 56 | Official leaderboard | |
| Structured Output Benchmark | 83.3% | #16 of 19 | Official leaderboard | |
| FACTS Parametric | 7.79% | #39 of 45 | Official leaderboard | |
| EvoArena SWE-Chain-Evo | 12% | #7 of 8 | Official leaderboard | |
| FACTS Benchmark Suite | 50.1% | #21 of 23 | Official leaderboard | |
| FACTS Parametric V2 | 6.91% | #26 of 27 | Official leaderboard | |
| HLE Extended | 3.80% | #30 of 31 | Official leaderboard | |
| AA-LCR Extended | 22.7% | #49 of 49 | Official leaderboard | |
| EvoArena Terminal-Bench-Evo | 24.5% | #8 of 8 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DTBench | 82.7% | #80 of 180 | Official leaderboard | |
| PHYSICS-Verified | 76.8% | #9 of 19 | Official leaderboard | |
| GPQA Diamond | 75.8% | #111 of 229 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 73.3% | #104 of 212 | Epoch AI run | |
| AIME 2026 | 89.2% | #28 of 38 | Lab model card | |
| ForecastBench | 0.15 | #64 of 82 | Official leaderboard | |
| Surface Evolver Bench | 30.6% | #21 of 25 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| StructEval | 77.5% | #5 of 28 | Official leaderboard | |
| Codeforces rating | 2,150 | #14 of 47 | Lab model card | |
| LiveCodeBench v6 | 80% | #32 of 90 | Lab model card | |
| SciCode | 43% | #24 of 49 | Lab model card | |
| ALE-Bench | 1,394 | #67 of 126 | Official leaderboard | |
| Code Arena | 1,370 | #94 of 128 | Official leaderboard | |
| WebDev Arena | 1,364 | #78 of 105 | arena.ai, via Epoch AI | |
| AndroidBench | 37.1% | #31 of 32 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 0% | #37 of 38 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Terminal-Bench Hard | 36% | #1 of 13 | Lab model card | |
| PACT | 1,557 | #8 of 26 | Official leaderboard | |
| TERMS-Bench | 0.64 | #4 of 11 | Official leaderboard | |
| GameHorizon | 66.3 | #20 of 37 | Official leaderboard | |
| tau2-bench | 76.9% | #46 of 79 | Lab model card | |
| PinchBench | 66.5% | #30 of 48 | Official leaderboard | |
| o11y-bench | 56.3% | #17 of 22 | Official leaderboard | |
| WildClawBench | 37.6% | #24 of 31 | Official leaderboard | |
| SnakeBench | 18 | #154 of 188 | Official leaderboard | |
| ClawProBench | 51.6% | #41 of 49 | Official leaderboard | |
| PinchBench 2.0 | 52.7% | #48 of 57 | Official leaderboard | |
| AutomationBench | 1.70% | #39 of 42 | Official leaderboard | |
| RuneBench | 1.57 | #73 of 76 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #51 of 53 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BRIDGE | 54.9 | #3 of 39 | Official leaderboard | |
| LEXam-hard | 37.4% | #6 of 15 | Official leaderboard | |
| MedXpertQA-MM | 61.3% | #9 of 21 | Lab model card | |
| FoodTruck Bench | $24,878 | #28 of 59 | Official leaderboard | |
| Vals Finance Agent | 50.8% | #27 of 49 | Official leaderboard | |
| LMBuild | 39.4% | #11 of 18 | Official leaderboard | |
| ProfBench | 44.4% | #47 of 72 | Official leaderboard | |
| MortgageTax | 61.4% | #57 of 87 | Official leaderboard | |
| 3DCodeArena | 921 | #10 of 14 | Official leaderboard | |
| RadLE 2.0 | 336.0 | #9 of 12 | Official leaderboard | |
| TRLawBench | 8.38 | #26 of 34 | Official leaderboard | |
| CaseLaw v2 | 52.6% | #44 of 53 | Official leaderboard | |
| APEX-1 | 48.1% | #69 of 71 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| InfographicVQA | 92% | #2 of 26 | Lab model card | |
| ChartMuseum | 71.1% | #4 of 21 | Official leaderboard | |
| Vision Arena | 1,265 | #32 of 136 | Official leaderboard | |
| VideoMME v2 (with subtitles) | 28.2% | #9 of 31 | Official leaderboard | |
| MATH-Vision (MathVision) | 85.6% | #17 of 51 | Lab model card | |
| MMMU-Pro | 76.9% | #32 of 90 | Lab model card | |
| CiteVQA | 20.2% | #7 of 16 | Official leaderboard | |
| ParseBench | 62.4% | #23 of 44 | Official leaderboard | |
| MineBench | 1,213 | #56 of 75 | Official leaderboard | |
| WorldBench | 49.7% | #6 of 8 | Official leaderboard | |
| PerceptionBench (Moonshot) | 40.7% | #13 of 16 | Official leaderboard | |
| 4DCodeBench | 0.19 | #16 of 16 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 81.7 | #54 of 185 | Official leaderboard | |
| BioTIER Permit | 99.5% | #31 of 69 | Official leaderboard | |
| Human Pathogen Capabilities Test | 53.6% | #40 of 86 | Official leaderboard | |
| BioTIER Refuse | 61.6% | #34 of 69 | Official leaderboard | |
| ABC Bench Fragment Design | 89.4% | #14 of 27 | Official leaderboard | |
| Molecular Biology Capabilities Test | 48.3% | #47 of 90 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 43.8% | #51 of 71 | Official leaderboard | |
| World-Class Bio | 33.7% | #55 of 74 | Official leaderboard | |
| ABC Bench Screening Evasion | 51.5% | #14 of 17 | Official leaderboard | |
| ABC Bench Liquid-Handling Robot | 19.3% | #31 of 34 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 55% | #2 of 81 | Official leaderboard | |
| Decision Index | 67.6 | #1 of 3 | Official leaderboard | |
| LMCA | 39.3% | #59 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 50.2 | #69 of 151 | Official leaderboard | |
| ACCoRD | 62.0 | #83 of 152 | Official leaderboard | |
| Political Science LLM Benchmark | 66.4% | #23 of 42 | Official leaderboard | |
| StudentBench | 12.7 | #9 of 11 | Official leaderboard | |
| SAGE v1.1 (Vals) | 82.8% | #24 of 29 | Official leaderboard | |
| LiveBench | 32.4% | #31 of 31 | Official leaderboard |
The Self-Reported Index fits one score to the 20 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 145.0 to 149.2. 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 | What differs |
|---|---|---|---|---|
| GPQA Diamond | 84.3% GPQA Diamond | 75.8% Epoch AI run, minimal | +8.5 pts | |
| LMArena (Elo) | 1,451 Arena Text Elo, as of June 19, 2026 (Table 4) | 1,452 Leaderboard | −1 | Google quotes Arena's June 19 rating, while the board reflects a later snapshot with more votes. |
| MATH-Vision (MathVision) | 85.6% MATH-Vision | 85.6% Leaderboard | Same | |
| MMMU-Pro | 76.9% MMMU Pro | 76.9% 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. 10 of them aren't in the launch post.