| Benchmark | Claude Sonnet 4.6 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 61.6 | 6.3 |
| AA Intelligence | 48.4 | 8.5 |
| AA-LCR | 74 | 24.7 |
| AA-Omniscience | 12.2 | -17.1 |
| Artificial Analysis Coding Index | 63 | 14.5 |
| critpt | 3.1 | 0 |
| gdpval | 54.8 | 0 |
| GPQA Diamond | 89.9 | 51.5 |
| harmbench | 24.2 | 62.7 |
| HLE | 49 | 4.2 |
| IFBench | 56.6 | 39 |
| mmmlu | 89.3 | 73.8 |
| OmniScience Accuracy | 40.9 | 23.2 |
| OmniScience Non-Hallucination | 51.6 | 49 |
| scicode | 47 | 29.9 |
| SWE-bench Pro | 58.1 | 11.2 |
| SWE-bench Verified | 79.6 | 24.5 |
| Terminal-Bench Hard | 59.1 | 6.8 |
| τ²-Bench Telecom (AA run) | 97.9 | 19 |
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.