| Benchmark | Gemini 3.5 Flash | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 70.4 | 32.1 |
| AA Intelligence | 52 | 26.1 |
| AA-LCR | 81 | 61.7 |
| AA-Omniscience | 21.2 | -50.8 |
| aime_2026 | 95 | 88.3 |
| arena_elo | 1477 | 1438 |
| arena_text_factuality | 1466 | 1445 |
| Artificial Analysis Coding Index | 70.1 | 39.3 |
| charxiv_rq | 84.2 | 69 |
| coding_arena_elo | 1490 | 1361 |
| critpt | 13.1 | 0 |
| gdpval | 57.8 | 25.7 |
| GDPval-AA v2 | 1357 | 807 |
| GPQA Diamond | 92.2 | 82.3 |
| HLE | 42.7 | 19.3 |
| hmmt_feb_2026 | 95.5 | 79 |
| IFBench | 76.3 | 72.4 |
| itbenchSre | 40.3 | 23.6 |
| MCP Atlas | 83.6 | 50 |
| MMMU-Pro | 84.3 | 73.8 |
| mrcr_v2_8needle_128k_average | 77.3 | 44.1 |
| OmniScience Accuracy | 51.4 | 19.1 |
| OmniScience Non-Hallucination | 38.2 | 13.6 |
| scicode | 53.1 | 40.3 |
| SWE-bench Pro | 55.1 | 13.8 |
| TauBench V3 - Banking | 32.2 | 12 |
| Terminal-Bench 2.0 | 76.2 | 34.2 |
| Terminal-Bench 2.1 | 78.7 | 39 |
| Terminal-Bench Hard | 46.2 | 25 |
| τ²-Bench Telecom (AA run) | 95.6 | 43.6 |
| τ³-Bench | 25.4 | 59 |
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.