| Benchmark | GPT-5 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 49.7 | 6.3 |
| AA Intelligence | 35.3 | 8.5 |
| AA-LCR | 76.3 | 24.7 |
| AA-Omniscience | -8.7 | -17.1 |
| aider_polyglot | 88 | 5.8 |
| AIR-Bench 2024 | 87.7 | 58.6 |
| Artificial Analysis Coding Index | 38.9 | 14.5 |
| critpt | 5.7 | 0 |
| Fortress | 17 | 20.6 |
| gdpval | 32.5 | 0 |
| GPQA Diamond | 87.3 | 51.5 |
| GSM8K | -1.8 | 96.8 |
| harmbench | 70 | 62.7 |
| HLE | 28.5 | 4.2 |
| humaneval | 93.4 | 89 |
| IFBench | 73.1 | 39 |
| LiveCodeBench | 85 | 30.1 |
| mmlu_redux | 95.3 | 86.2 |
| MMLU-Pro | 87.5 | 73.4 |
| OmniScience Accuracy | 40.3 | 23.2 |
| OmniScience Non-Hallucination | 21 | 49 |
| scicode | 43 | 29.9 |
| simplebench | 56.7 | 23 |
| SWE-bench Pro | 41.8 | 11.2 |
| SWE-bench Verified | 74.9 | 24.5 |
| Terminal-Bench Hard | 37.9 | 6.8 |
| τ²-Bench Telecom (AA run) | 86.5 | 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.