| Benchmark | Claude Opus 4.8 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 49.4 | 6.3 |
| AA Intelligence | 57.3 | 8.5 |
| AA-LCR | 73 | 24.7 |
| AA-Omniscience | 28.8 | -17.1 |
| Artificial Analysis Coding Index | 74.3 | 14.5 |
| critpt | 20.9 | 0 |
| Fortress | 18.2 | 20.6 |
| gdpval | 54.2 | 0 |
| GPQA Diamond | 93.6 | 51.5 |
| HLE | 57.9 | 4.2 |
| IFBench | 62.2 | 39 |
| longbench_v2 | 69.1 | 36.1 |
| OmniScience Accuracy | 48.8 | 23.2 |
| OmniScience Non-Hallucination | 60.7 | 49 |
| scicode | 53.5 | 29.9 |
| simplebench | 64.8 | 23 |
| SWE-bench Pro | 69.2 | 11.2 |
| SWE-bench Verified | 88.6 | 24.5 |
| Terminal-Bench Hard | 58.3 | 6.8 |
| τ²-Bench Telecom (AA run) | 94.4 | 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.