| Benchmark | o4-mini | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 26.1 | 10 |
| AA-LCR | 60 | 20.3 |
| AA-Omniscience | -35.7 | -52.2 |
| aider_polyglot | 72 | 7.6 |
| AIR-Bench 2024 | 78.5 | 59 |
| anthropic_red_team | 98.2 | 99.6 |
| Artificial Analysis Coding Index | 25.6 | 11.9 |
| bbq | 94 | 95.4 |
| critpt | 0.6 | 0 |
| Fortress | 21.5 | 56.4 |
| GPQA Diamond | 81.4 | 49.1 |
| GSM8K | 0 | 95.8 |
| harmbench | 97 | 72.8 |
| HLE | 16.5 | 4.2 |
| IFBench | 68.7 | 36.9 |
| LiveCodeBench | 84.5 | 55.5 |
| OmniScience Accuracy | 24.8 | 17.6 |
| OmniScience Non-Hallucination | 19.5 | 15.3 |
| scicode | 46.5 | 26.7 |
| simple_safety_tests | 100 | 100 |
| SWE-bench Verified | 68.1 | 23.8 |
| Terminal-Bench Hard | 15.2 | 4.5 |
| xstest | 97.4 | 97.9 |
| τ²-Bench Telecom (AA run) | 55.6 | 34.5 |
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.