| Benchmark | Gemini 2.5 Pro | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 7.2 | 1 |
| AA Intelligence | 27 | 7 |
| aider_polyglot | 82.2 | 3.6 |
| AIR-Bench 2024 | 73.6 | 56.3 |
| anthropic_red_team | 99.5 | 98.3 |
| ARC-AGI-2 | 4.9 | 0 |
| Artificial Analysis Coding Index | 46.7 | 11.4 |
| bbq | 96.4 | 88.2 |
| Fortress | 54.9 | 48.1 |
| gdpval | 8.5 | 0 |
| GPQA Diamond | 84.4 | 42.6 |
| harmbench | 65.4 | 84.9 |
| HLE | 22.5 | 4.2 |
| IFBench | 49 | 31 |
| LiveCodeBench | 82.7 | 35.5 |
| livecodebench_easy | 98.8 | 81.7 |
| livecodebench_hard | 59.4 | 4.3 |
| livecodebench_medium | 90.6 | 25.2 |
| MathVista | 83.9 | 56.7 |
| MMMU | 83.9 | 59.4 |
| MMMU-Pro | 74.9 | 41.5 |
| scicode | 43 | 22.9 |
| simple_safety_tests | 97 | 97.8 |
| simplebench | 62.4 | 10.7 |
| simpleqa | 50.8 | 9.9 |
| SWE-bench Verified | 67.2 | 8.7 |
| TauBench V3 - Banking | 9.7 | 2.9 |
| Terminal-Bench 2.1 | 28.5 | 5.6 |
| Video-MME | 84.8 | 68.9 |
| xstest | 98.7 | 96 |
| τ³-Bench | 9.3 | 2.9 |
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.