| Benchmark | GPT-5.1 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 32.2 | 6.3 |
| AA Intelligence | 37.5 | 8.5 |
| AA-LCR | 76.7 | 24.7 |
| AA-Omniscience | 5.4 | -17.1 |
| Artificial Analysis Coding Index | 49.4 | 14.5 |
| critpt | 4.9 | 0 |
| Fortress | 25.7 | 20.6 |
| gdpval | 25 | 0 |
| GPQA Diamond | 88.1 | 51.5 |
| HLE | 28.5 | 4.2 |
| IFBench | 72.9 | 39 |
| MMLU-Pro | 87 | 73.4 |
| mmmlu | 91 | 73.8 |
| OmniScience Accuracy | 37.7 | 23.2 |
| OmniScience Non-Hallucination | 48.1 | 49 |
| scicode | 43.3 | 29.9 |
| simplebench | 53.2 | 23 |
| SWE-bench Verified | 76.3 | 24.5 |
| Terminal-Bench Hard | 45.5 | 6.8 |
| τ²-Bench Telecom (AA run) | 81.9 | 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.