| Benchmark | Gemini 3.6 Flash | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 40.5 | 32.1 |
| AA Intelligence | 52 | 26.1 |
| AA-LCR | 79 | 61.7 |
| AA-Omniscience | 22.1 | -50.8 |
| aime_2026 | 96.7 | 88.3 |
| arena_elo | 1481 | 1438 |
| arena_text_factuality | 1470 | 1445 |
| Artificial Analysis Coding Index | 69.2 | 39.3 |
| charxiv_rq | 85.2 | 69 |
| coding_arena_elo | 1538 | 1361 |
| critpt | 10.6 | 0 |
| gdpval | 46.1 | 25.7 |
| GDPval-AA v2 | 1421 | 807 |
| GPQA Diamond | 92.8 | 82.3 |
| HLE | 40.8 | 19.3 |
| hmmt_feb_2026 | 89.4 | 79 |
| MMMU-Pro | 83.2 | 73.8 |
| mrcr_v2_8needle_128k_average | 91.8 | 44.1 |
| OmniScience Accuracy | 50 | 19.1 |
| OmniScience Non-Hallucination | 44.4 | 13.6 |
| scicode | 52.7 | 40.3 |
| SWE-bench Pro | 58.7 | 13.8 |
| TauBench V3 - Banking | 29.9 | 12 |
| Terminal-Bench 2.1 | 78 | 39 |
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.