| Benchmark | GPT-5.6 Sol | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 57.8 | 6.3 |
| AA Intelligence | 61 | 8.5 |
| AA-LCR | 77.7 | 24.7 |
| AA-Omniscience | 22 | -17.1 |
| Artificial Analysis Coding Index | 78.3 | 14.5 |
| critpt | 32.3 | 0 |
| gdpval | 61.1 | 0 |
| GPQA Diamond | 94.6 | 51.5 |
| HLE | 49.5 | 4.2 |
| IFBench | 72.7 | 39 |
| longbench_v2 | 67.1 | 36.1 |
| OmniScience Accuracy | 59.4 | 23.2 |
| OmniScience Non-Hallucination | 10.6 | 49 |
| scicode | 56.9 | 29.9 |
| simplebench | 64.8 | 23 |
| SWE-bench Pro | 64.6 | 11.2 |
| Terminal-Bench Hard | 65.9 | 6.8 |
| τ²-Bench Telecom (AA run) | 85.1 | 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.