| Benchmark | Gemini 3 Flash Preview | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 49.7 | 32.1 |
| AA Intelligence | 38.7 | 26.1 |
| AA-LCR | 73 | 61.7 |
| AA-Omniscience | 10.1 | -50.8 |
| AIME 2025 | 99.7 | 90 |
| aime_2026 | 96.7 | 88.3 |
| arena_elo | 1473 | 1438 |
| arena_text_factuality | 1452 | 1445 |
| arena_vision | 1260 | 1238 |
| Artificial Analysis Coding Index | 42.6 | 39.3 |
| charxiv_rq | 80.3 | 69 |
| coding_arena_elo | 1387 | 1361 |
| critpt | 8.6 | 0 |
| gdpval | 35.2 | 25.7 |
| GPQA Diamond | 90.4 | 82.3 |
| HLE | 43.5 | 19.3 |
| hmmt_feb_2025 | 97.5 | 91.7 |
| hmmt_feb_2026 | 89.4 | 79 |
| hmmt_nov_2025 | 93.3 | 87.5 |
| IFBench | 78 | 72.4 |
| MCP Atlas | 62 | 50 |
| mmmlu | 91.8 | 86.3 |
| MMMU-Pro | 81.2 | 73.8 |
| mrcr_v2_8needle_128k_average | 67.2 | 44.1 |
| OmniDocBench 1.5 | 12.1 | 74.4 |
| OmniScience Accuracy | 53.4 | 19.1 |
| OmniScience Non-Hallucination | 7.6 | 13.6 |
| scicode | 50.6 | 40.3 |
| SWE-bench Multilingual | 72.7 | 43.4 |
| SWE-bench Pro | 49.6 | 13.8 |
| SWE-bench Verified | 78 | 57.4 |
| TauBench V3 - Banking | 20.8 | 12 |
| Terminal-Bench 2.0 | 47.6 | 34.2 |
| Terminal-Bench 2.1 | 58 | 39 |
| Terminal-Bench Hard | 38.6 | 25 |
| τ²-Bench Telecom (AA run) | 80.4 | 43.6 |
| τ³-Bench | 17.5 | 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.