| Benchmark | Gemma 4 E2B | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 7.4 | 60.2 |
| AA Intelligence | 10 | 43.3 |
| AA-LCR | 17 | 79.3 |
| AA-Omniscience | -23.6 | 0.3 |
| aime_2026 | 37.5 | 98.3 |
| AIME25 no tools | 26.3 | 98 |
| Artificial Analysis Coding Index | 8.3 | 48.7 |
| critpt | 0 | 11.6 |
| gdpval | 0 | 48.3 |
| GPQA Diamond | 43.4 | 92.4 |
| HLE | 4.8 | 45.5 |
| IFBench | 38 | 75.4 |
| mathvision | 52.4 | 83 |
| MathVista | 62.1 | 82.8 |
| MMLU-Pro | 60 | 87 |
| mmmlu | 67.4 | 89.6 |
| MMMU-Pro | 44.6 | 79.5 |
| mrcr_v2_8needle_128k_average | 19.1 | 83.8 |
| OCRBench | 0.7 | 80.7 |
| ocrbench_v2 | 44.7 | 50.5 |
| OmniScience Accuracy | 8.2 | 44.3 |
| OmniScience Non-Hallucination | 67.6 | 38.4 |
| scicode | 20.9 | 52.1 |
| Tau2 retail | 18.9 | 82 |
| TauBench V3 - Banking | 4.5 | 11.1 |
| Terminal-Bench Hard | 3 | 62.2 |
| τ²-Bench | 24.5 | 85.5 |
| τ²-Bench Telecom (AA run) | 22.2 | 98.7 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.