| Benchmark | Gemini 1.5 Pro | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 10 | 10 |
| aider_polyglot | 16.9 | 7.6 |
| AIR-Bench 2024 | 82.8 | 59 |
| anthropic_red_team | 99.9 | 99.6 |
| Arena Hard | 85.3 | 81.2 |
| Artificial Analysis Coding Index | 23.6 | 11.9 |
| bbh | 89.2 | 79.8 |
| bbq | 94.5 | 95.4 |
| Chinese SimpleQA (C-SimpleQA) | 59.4 | 52.2 |
| DROP | 74.9 | 76.7 |
| DROP (F1) | 89.2 | 85 |
| Fortress | 53.9 | 56.4 |
| GPQA Diamond | 59.1 | 49.1 |
| GSM8K | 95.2 | 95.8 |
| harmbench | 79.9 | 72.8 |
| hellaswag | 93.3 | 84.8 |
| HLE | 4.9 | 4.2 |
| humaneval | 86.6 | 86.6 |
| ifeval | 89.4 | 87.2 |
| IFEval (avg) | 89.4 | 87.2 |
| LiveCodeBench | 30.5 | 55.5 |
| MATH | 92 | 88.4 |
| MBPP+ (EvalPlus-augmented) | 75.4 | 77 |
| mgsm | 87.5 | 87.3 |
| mmlu | 86.8 | 86.1 |
| MMLU-Pro | 75.8 | 71.6 |
| narrativeqa | 78.3 | 74.5 |
| naturalquestions_closedbook | 45.5 | 35.9 |
| OpenBookQA | 95.2 | 96.2 |
| scicode | 29.5 | 26.7 |
| simple_safety_tests | 97.5 | 100 |
| simpleqa | 23.4 | 10.3 |
| SWE-bench Verified | 34.2 | 23.8 |
| xstest | 98.8 | 97.9 |
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.