| Benchmark | DeepSeek-V3 | Gemini 1.5 Pro |
|---|---|---|
| AA Intelligence | 15.4 | 10 |
| aider_polyglot | 55.1 | 16.9 |
| AIR-Bench 2024 | 40.8 | 82.8 |
| anthropic_red_team | 97.1 | 99.9 |
| Arena Hard | 91.4 | 85.3 |
| Artificial Analysis Coding Index | 23 | 23.6 |
| bbh | 87.5 | 89.2 |
| bbq | 96.7 | 94.5 |
| Chinese SimpleQA (C-SimpleQA) | 68 | 59.4 |
| DROP | 91.6 | 74.9 |
| DROP (F1) | 91 | 89.2 |
| GPQA Diamond | 68.4 | 59.1 |
| GSM8K | 96.7 | 95.2 |
| harmbench | 49.7 | 79.9 |
| hellaswag | 88.9 | 93.3 |
| HLE | 5.2 | 4.9 |
| humaneval | 92.1 | 86.6 |
| ifeval | 87.3 | 89.4 |
| IFEval (avg) | 87.3 | 89.4 |
| LiveCodeBench | 49.6 | 30.5 |
| MATH | 91.2 | 92 |
| MBPP+ (EvalPlus-augmented) | 78.8 | 75.4 |
| mgsm | 79.8 | 87.5 |
| mmlu | 89.4 | 86.8 |
| MMLU-Pro | 81.2 | 75.8 |
| narrativeqa | 79.6 | 78.3 |
| naturalquestions_closedbook | 46.7 | 45.5 |
| OpenBookQA | 95.4 | 95.2 |
| scicode | 35.8 | 29.5 |
| simple_safety_tests | 95.3 | 97.5 |
| simplebench | 27.2 | 27.1 |
| simpleqa | 27.7 | 23.4 |
| SWE-bench Verified | 42 | 34.2 |
| xstest | 97.1 | 98.8 |
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.