Gemma 4 26B A4B places in the top three on 2 benchmarks.
Google DeepMind Apr 2026 Open 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 | |
|---|---|---|---|---|
| IFEval | 98.5% | #2 of 135 | Lab model card | |
| Vectara Hallucination Leaderboard (2025 dataset) | 5.20% | #14 of 104 | Official leaderboard | |
| SEA-HELM | 74.8 | #14 of 71 | Official leaderboard | |
| FACTS Grounding v2 | 77.8% | #5 of 25 | Official leaderboard | |
| BBEH micro average | 64.8% | #3 of 15 | Lab model card | |
| LMArena (Elo) | 1,437 | #80 of 335 | Official leaderboard | |
| MMLU-Pro | 82.6% | #56 of 167 | Lab model card | |
| MMMLU | 86.3% | #24 of 71 | Lab model card | |
| EQ-Bench 3 | 1,455 | #27 of 78 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 10667.0 | #34 of 77 | Official leaderboard | |
| IFBench | 72% | #36 of 80 | Lab model card | |
| RULER | 89.8% | #21 of 42 | Lab model card | |
| QIMMA | 51.8% | #13 of 25 | Official leaderboard | |
| Creative Writing v3 | 65.8 | #66 of 120 | Official leaderboard | |
| Judgemark v4 | 53.0 | #25 of 40 | Official leaderboard | |
| Global MMLU Lite | 85.7% | #28 of 43 | Official leaderboard | |
| GraphWalks | 72.6% | #18 of 27 | Lab model card | |
| Longform Creative Writing (EQ-Bench) | 50.7 | #82 of 119 | Official leaderboard | |
| BullshitBench V2 | 21% | #91 of 129 | Official leaderboard | |
| LOFT | 66.3% | #6 of 8 | Lab model card | |
| Humanity's Last Exam | 17.2% | #118 of 154 | Lab model card | |
| FACTS Multimodal | 39.5% | #18 of 23 | Official leaderboard | |
| MRCR v2 | 44.1% | #24 of 30 | Lab model card | |
| BullshitBench | 14.6% | #89 of 107 | Official leaderboard | |
| SimpleQA Verified | 9.05% | #49 of 56 | Official leaderboard | |
| FACTS Search | 56.7% | #28 of 31 | Official leaderboard | |
| FACTS Parametric | 5.69% | #41 of 45 | Official leaderboard | |
| FACTS Benchmark Suite | 47.4% | #22 of 23 | Official leaderboard | |
| FACTS Parametric V2 | 5.66% | #27 of 27 | Official leaderboard | |
| FACTS Search V2 | 60.3% | #21 of 21 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 82.2% | #94 of 212 | Epoch AI run | |
| GPQA Diamond | 73.2% | #113 of 229 | Epoch AI run | |
| DTBench | 74.9% | #107 of 180 | Official leaderboard | |
| AIME 2026 | 88.3% | #29 of 38 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 77.1% | #39 of 90 | Lab model card | |
| SciCode | 40% | #30 of 49 | Lab model card | |
| WebDev Arena | 1,361 | #79 of 105 | arena.ai, via Epoch AI | |
| Code Arena | 1,358 | #98 of 128 | Official leaderboard | |
| Codeforces rating | 1,718 | #36 of 47 | Lab model card | |
| StructEval | 68.9% | #22 of 28 | Official leaderboard | |
| Chart Arena | 1,322 | #15 of 15 | Official leaderboard | |
| ScarfBench | 24.4% | #11 of 11 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PinchBench | 74.1% | #13 of 48 | Official leaderboard | |
| SnakeBench | 24 | #98 of 188 | Official leaderboard | |
| Terminal-Bench Hard | 14% | #9 of 13 | Lab model card | |
| PinchBench 2.0 | 56.4% | #43 of 57 | Official leaderboard | |
| tau2-bench | 68.2% | #61 of 79 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedXpertQA-MM | 58.1% | #10 of 21 | Lab model card | |
| FoodTruck Bench | $4,386 | #42 of 59 | Official leaderboard | |
| TRLawBench | 8.11 | #27 of 34 | Official leaderboard | |
| Physical-Engineering's Last Exam | 0.04 | #16 of 18 | Official leaderboard | |
| 3DCodeArena | 828 | #13 of 14 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| InfographicVQA | 89.3% | #5 of 26 | Lab model card | |
| ChartMuseum | 70.6% | #5 of 21 | Official leaderboard | |
| Vision Arena | 1,242 | #53 of 136 | Official leaderboard | |
| MATH-Vision (MathVision) | 82.4% | #20 of 51 | Lab model card | |
| MMMU-Pro | 73.8% | #46 of 90 | Lab model card | |
| CiteVQA | 6.20% | #16 of 16 | Official leaderboard | |
| OVO-S-Bench | 42.6% | #15 of 15 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMCA | 29.7% | #90 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 40.3 | #99 of 151 | Official leaderboard | |
| ACCoRD | 54.3 | #109 of 152 | 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 138.6 to 143.2. 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 | 82.3% GPQA Diamond | 73.2% Epoch AI run, minimal | +9.1 pts | |
| LMArena (Elo) | 1,438 Arena Text Elo, as of June 19, 2026 (Table 4) | 1,437 Leaderboard | +1 | The lab quotes a rounded June 19 Arena rating, while the board uses the November 8 rating. |
| MATH-Vision (MathVision) | 82.4% MATH-Vision | 82.4% Leaderboard | Same | |
| MMMU-Pro | 73.8% MMMU Pro | 73.8% 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.