| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 1.6 | 1 |
| AA Intelligence | 15.4 | 7 |
| aider_polyglot | 55.1 | 3.6 |
| AIR-Bench 2024 | 40.8 | 56.3 |
| anthropic_red_team | 97.1 | 98.3 |
| Artificial Analysis Coding Index | 23 | 11.4 |
| bbq | 96.7 | 88.2 |
| DROP | 91.6 | 79.7 |
| gdpval | 0 | 0 |
| GPQA Diamond | 68.4 | 42.6 |
| GSM8K | 96.7 | 84.3 |
| harmbench | 49.7 | 84.9 |
| HLE | 5.2 | 4.2 |
| humaneval | 92.1 | 87.2 |
| IFBench | 41 | 31 |
| LiveCodeBench | 49.6 | 35.5 |
| livecodebench_easy | 83.3 | 81.7 |
| livecodebench_hard | 14.2 | 4.3 |
| livecodebench_medium | 53.5 | 25.2 |
| MATH | 91.2 | 80.2 |
| mgsm | 79.8 | 87 |
| mmlu | 89.4 | 81.8 |
| narrativeqa | 79.6 | 76.8 |
| naturalquestions_closedbook | 46.7 | 38.5 |
| OpenBookQA | 95.4 | 92 |
| scicode | 35.8 | 22.9 |
| simple_safety_tests | 95.3 | 97.8 |
| simplebench | 27.2 | 10.7 |
| simpleqa | 27.7 | 9.9 |
| SWE-bench Verified | 42 | 8.7 |
| TauBench V3 - Banking | 4.7 | 2.9 |
| Terminal-Bench 2.1 | 16.9 | 5.6 |
| xstest | 97.1 | 96 |
| τ³-Bench | 4.7 | 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.