| Benchmark | Llama 3.1 Instruct 405B | o4-mini |
|---|---|---|
| AA Agentic Index | 6.3 | 36.1 |
| AA Intelligence | 8.5 | 26.1 |
| AA-LCR | 24.7 | 60 |
| AA-Omniscience | -17.1 | -35.7 |
| aider_polyglot | 5.8 | 72 |
| AIR-Bench 2024 | 58.6 | 78.5 |
| anthropic_red_team | 96.5 | 98.2 |
| Artificial Analysis Coding Index | 14.5 | 25.6 |
| bbq | 94.5 | 94 |
| critpt | 0 | 0.6 |
| Fortress | 20.6 | 21.5 |
| gdpval | 0 | 25.4 |
| GPQA Diamond | 51.5 | 81.4 |
| GSM8K | 96.8 | 0 |
| harmbench | 62.7 | 97 |
| HLE | 4.2 | 16.5 |
| IFBench | 39 | 68.7 |
| LiveCodeBench | 30.1 | 84.5 |
| OmniScience Accuracy | 23.2 | 24.8 |
| OmniScience Non-Hallucination | 49 | 19.5 |
| scicode | 29.9 | 46.5 |
| simple_safety_tests | 98.8 | 100 |
| simplebench | 23 | 38.7 |
| SWE-bench Verified | 24.5 | 68.1 |
| Terminal-Bench Hard | 6.8 | 15.2 |
| xstest | 95.9 | 97.4 |
| τ²-Bench Telecom (AA run) | 19 | 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.