| Benchmark | Llama 3.1 Instruct 405B | Muse Spark |
|---|---|---|
| AA Agentic Index | 6.3 | 28.7 |
| AA Intelligence | 8.5 | 44.3 |
| AA-LCR | 24.7 | 77 |
| AA-Omniscience | -17.1 | 7.2 |
| Artificial Analysis Coding Index | 14.5 | 58.6 |
| critpt | 0 | 11.3 |
| Fortress | 20.6 | 20.2 |
| gdpval | 0 | 32.3 |
| GPQA Diamond | 51.5 | 89.5 |
| HLE | 4.2 | 58.4 |
| IFBench | 39 | 75.9 |
| OmniScience Accuracy | 23.2 | 49.6 |
| OmniScience Non-Hallucination | 49 | 26.8 |
| scicode | 29.9 | 51.5 |
| SWE-bench Pro | 11.2 | 55 |
| SWE-bench Verified | 24.5 | 77.4 |
| Terminal-Bench Hard | 6.8 | 45.5 |
| τ²-Bench Telecom (AA run) | 19 | 91.5 |
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.