DiffusionGemma 26B A4B ranks first on 2 of the 22 benchmarks with five or more models scored.
Google DeepMind Jun 2026 Open 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 | |
|---|---|---|---|---|
| IFEval | 97.4% | #3 of 135 | Lab model card | |
| BBEH micro average | 47.6% | #7 of 15 | Lab model card | |
| MGSM | 84.8% | #27 of 57 | Lab model card | |
| MMMLU | 81.5% | #40 of 71 | Lab model card | |
| MMLU-Pro | 77.6% | #95 of 167 | Lab model card | |
| Humanity's Last Exam | 11.9% | #132 of 154 | Lab model card | |
| MRCR v2 | 32% | #26 of 30 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GSM8K | 96.3% | #5 of 92 | Lab model card | |
| HiddenMath | 80.6% | #1 of 11 | Lab model card | |
| GPQA Diamond | 73.2% | #160 of 302 | Lab model card | |
| PutnamBench | 67.4% | #4 of 7 | Lab model card | |
| AIME 2026 | 69.1% | #34 of 38 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HumanEval | 94.5% | #5 of 101 | Lab model card | |
| Natural2Code | 94% | #1 of 9 | Lab model card | |
| LBPP (Less Basic Python Problems) | 81% | #1 of 2 | Lab model card | |
| BigCodeBench | 46% | #9 of 14 | Lab model card | |
| LiveCodeBench v6 | 69.1% | #58 of 90 | Lab model card | |
| Codeforces rating | 1,429 | #42 of 47 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 56.2% | #69 of 79 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedXpertQA-MM | 49% | #13 of 21 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH-Vision (MathVision) | 70.5% | #31 of 51 | Lab model card | |
| MMMU-Pro | 54.3% | #74 of 90 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 70.2 | #101 of 185 | Official leaderboard |
The Self-Reported Index fits one score to the 19 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 131.7 to 136.1. How we compute it.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 9 of them aren't in the launch post.