| Benchmark | Llama 3.1 Instruct 405B | Opus 5 |
|---|---|---|
| AA Agentic Index | 6.3 | 59.2 |
| AA Intelligence | 8.5 | 63.1 |
| AA-LCR | 24.7 | 78.7 |
| AA-Omniscience | -17.1 | 37.1 |
| Artificial Analysis Coding Index | 14.5 | 78 |
| critpt | 0 | 29.1 |
| gdpval | 0 | 67.2 |
| GPQA Diamond | 51.5 | 93.7 |
| HLE | 4.2 | 64.7 |
| OmniScience Accuracy | 23.2 | 60.9 |
| OmniScience Non-Hallucination | 49 | 40.5 |
| scicode | 29.9 | 55.7 |
| simplebench | 23 | 80.6 |
| SWE-bench Pro | 11.2 | 79.2 |
| SWE-bench Verified | 24.5 | 96 |
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.