Gemma 4 E2B places in the top three on MSEB.
Google DeepMind Mar 2026 Open weights Small
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 | 94.6% | #9 of 135 | Lab model card | |
| Global PIQA | 72.1% | #20 of 44 | Official leaderboard | |
| AgentIF (2026-10 setting) | 61% | #15 of 24 | Official leaderboard | |
| BOUQuET | 26.0 | #12 of 18 | Official leaderboard | |
| IFStruct | 64.8% | #7 of 10 | Official leaderboard | |
| BBEH micro average | 21.9% | #11 of 15 | Lab model card | |
| SEA-HELM | 58.8 | #53 of 71 | Official leaderboard | |
| IFMTBench | 58.7% | #6 of 8 | Official leaderboard | |
| MMMLU | 67.4% | #60 of 71 | Lab model card | |
| MMLU-Pro | 60% | #145 of 167 | Lab model card | |
| RULER | 70.4% | #38 of 42 | Lab model card | |
| IFBench | 38% | #73 of 80 | Lab model card | |
| QIMMA | 38.1% | #23 of 25 | Official leaderboard | |
| MultiChallenge | 36% | #39 of 42 | Official leaderboard | |
| MRCR v2 | 19.1% | #30 of 30 | Lab model card | |
| GraphWalks | 4.10% | #27 of 27 | Lab model card | |
| LOFT | 50.5% | #8 of 8 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PHYSICS-Verified | 20.2% | #14 of 19 | Official leaderboard | |
| GPQA Diamond | 43.4% | #262 of 302 | Lab model card | |
| AIME 2026 | 37.5% | #38 of 38 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 44% | #86 of 90 | Lab model card | |
| SciCode | 21% | #47 of 49 | Lab model card | |
| Codeforces rating | 633 | #46 of 47 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 24.5% | #78 of 79 | Lab model card | |
| Terminal-Bench Hard | 3.00% | #13 of 13 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BRIDGE | 42.3 | #30 of 39 | Official leaderboard | |
| MedXpertQA-MM | 23.5% | #20 of 21 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MSEB | 17.9 | #3 of 9 | Official leaderboard | |
| OmniDocBench v1.5 | 0.29 | #6 of 9 | Lab model card | |
| ExtractBench | 51.9% | #23 of 27 | Official leaderboard | |
| MATH-Vision (MathVision) | 52.4% | #45 of 51 | Lab model card | |
| InfographicVQA | 63.9% | #23 of 26 | Lab model card | |
| MMMU-Pro | 44.2% | #82 of 90 | Lab model card | |
| CoVoST2 | 35.4 | #13 of 14 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 44.0 | #173 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 113.7 to 119.3. 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.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 8 of them aren't in the launch post.