| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| AA Agentic Index | 1.6 | 36.1 |
| AA Intelligence | 15.4 | 31.1 |
| AA-LCR | 41.3 | 73.3 |
| AA-Omniscience | -37.6 | -15.3 |
| aider_polyglot | 55.1 | 81.3 |
| AIME 2024 | 59.4 | 91.6 |
| AIME 2025 | 51.3 | 89.2 |
| AIR-Bench 2024 | 40.8 | 84.5 |
| anthropic_red_team | 97.1 | 98.3 |
| Artificial Analysis Coding Index | 23 | 38.4 |
| bbq | 96.7 | 97.9 |
| critpt | 0 | 1.1 |
| gdpval | 0 | 12.8 |
| GPQA Diamond | 68.4 | 83.3 |
| harmbench | 49.7 | 98.4 |
| HLE | 5.2 | 20.3 |
| hmmt_feb_2025 | 29.2 | 77.5 |
| IFBench | 41 | 71.4 |
| LiveCodeBench | 49.6 | 84.7 |
| livecodebench_easy | 83.3 | 99.1 |
| livecodebench_hard | 14.2 | 66 |
| livecodebench_medium | 53.5 | 89.8 |
| longbench_v2 | 48.7 | 58.8 |
| math_500_em | 94 | 98.1 |
| MMLU-Pro | 81.2 | 85 |
| multichallenge | 31.4 | 60.4 |
| OmniScience Accuracy | 25.4 | 38.6 |
| OmniScience Non-Hallucination | 23.3 | 12.9 |
| scicode | 35.8 | 41 |
| simple_safety_tests | 95.3 | 99 |
| simplebench | 27.2 | 53.1 |
| simpleqa | 27.7 | 49.4 |
| SWE-bench Verified | 42 | 69.1 |
| Tau2 airline | 39 | 64.8 |
| Tau2 retail | 69.1 | 80.2 |
| Terminal-Bench Hard | 15.2 | 37.1 |
| xstest | 97.1 | 97.3 |
| ZebraLogic | 84 | 95.8 |
| τ²-Bench Telecom (AA run) | 47.1 | 80.7 |
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.