| Benchmark | DeepSeek-V3 | o4-mini |
|---|---|---|
| AA Agentic Index | 1.6 | 36.1 |
| AA Intelligence | 15.4 | 26.1 |
| AA-LCR | 41.3 | 60 |
| AA-Omniscience | -37.6 | -35.7 |
| aider_polyglot | 55.1 | 72 |
| AIME 2025 | 51.3 | 92.7 |
| AIR-Bench 2024 | 40.8 | 78.5 |
| anthropic_red_team | 97.1 | 98.2 |
| Artificial Analysis Coding Index | 23 | 25.6 |
| bbq | 96.7 | 94 |
| critpt | 0 | 0.6 |
| gdpval | 0 | 25.4 |
| GPQA Diamond | 68.4 | 81.4 |
| GSM8K | 96.7 | 0 |
| harmbench | 49.7 | 97 |
| HLE | 5.2 | 16.5 |
| IFBench | 41 | 68.7 |
| LiveCodeBench | 49.6 | 84.5 |
| livecodebench_easy | 83.3 | 98.8 |
| livecodebench_hard | 14.2 | 62.9 |
| livecodebench_medium | 53.5 | 92.2 |
| mmlu_prox | 70.5 | 69.3 |
| multichallenge | 31.4 | 44.9 |
| OmniScience Accuracy | 25.4 | 24.8 |
| OmniScience Non-Hallucination | 23.3 | 19.5 |
| scicode | 35.8 | 46.5 |
| simple_safety_tests | 95.3 | 100 |
| simplebench | 27.2 | 38.7 |
| SWE-bench Verified | 42 | 68.1 |
| Terminal-Bench Hard | 15.2 | 15.2 |
| vectara_answer_rate | 97.5 | 99.2 |
| vectara_avg_summary_length | 81.7 | 130.9 |
| vectara_factual_consistency | 93.9 | 81.4 |
| vectara_hallucination_rate ↓ | 6.1 | 18.6 |
| xstest | 97.1 | 97.4 |
| τ²-Bench Telecom (AA run) | 47.1 | 55.6 |
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.