| Benchmark | Gemini 2.5 Flash | Gemma 4 26B A4B |
|---|---|---|
| AA Agentic Index | 18.8 | 32.1 |
| AA Intelligence | 17.7 | 26.1 |
| AA-LCR | 65.7 | 61.7 |
| AA-Omniscience | -29.8 | -50.8 |
| AIME 2025 | 72 | 90 |
| arena_text_factuality | 1437 | 1445 |
| arena_vision | 1214 | 1238 |
| Artificial Analysis Coding Index | 22.2 | 39.3 |
| critpt | 1.4 | 0 |
| gdpval | 11.9 | 25.7 |
| GPQA Diamond | 82.8 | 82.3 |
| HLE | 12.1 | 19.3 |
| hmmt_feb_2025 | 64.2 | 91.7 |
| IFBench | 50.3 | 72.4 |
| LiveCodeBench | 76.2 | 79.8 |
| MMMU | 79.7 | 78.4 |
| MMMU-Pro | 69.1 | 73.8 |
| OmniScience Accuracy | 26.1 | 19.1 |
| OmniScience Non-Hallucination | 24.7 | 13.6 |
| scicode | 35.9 | 40.3 |
| SWE-bench Verified | 60.4 | 57.4 |
| Terminal-Bench Hard | 13.6 | 25 |
| τ²-Bench Telecom (AA run) | 31.6 | 43.6 |
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.