| Benchmark | DeepSeek-R1 | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 3.1 | 1 |
| AA Intelligence | 18.6 | 7 |
| aider_polyglot | 71.6 | 3.6 |
| AIR-Bench 2024 | 52.9 | 56.3 |
| anthropic_red_team | 99.1 | 98.3 |
| ARC-AGI-2 | 1.3 | 0 |
| Artificial Analysis Coding Index | 24.6 | 11.4 |
| bbq | 96.6 | 88.2 |
| Fortress | 74.4 | 48.1 |
| gdpval | 1.5 | 0 |
| GPQA Diamond | 81 | 42.6 |
| harmbench | 54.6 | 84.9 |
| HLE | 17.7 | 4.2 |
| IFBench | 39 | 31 |
| LiveCodeBench | 84.4 | 35.5 |
| livecodebench_easy | 99.2 | 81.7 |
| livecodebench_hard | 63.6 | 4.3 |
| livecodebench_medium | 90.9 | 25.2 |
| MATH | 97.3 | 80.2 |
| mmlu | 92.9 | 81.8 |
| scicode | 35.7 | 22.9 |
| simple_safety_tests | 98.3 | 97.8 |
| simplebench | 40.8 | 10.7 |
| simpleqa | 92.3 | 9.9 |
| SWE-bench Verified | 57.6 | 8.7 |
| TauBench V3 - Banking | 6.4 | 2.9 |
| Terminal-Bench 2.1 | 19.1 | 5.6 |
| xstest | 98.8 | 96 |
| τ³-Bench | 6.4 | 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.