| Benchmark | GPT-5.2 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 60.2 | 6.3 |
| AA Intelligence | 43.3 | 8.5 |
| AA-LCR | 79.3 | 24.7 |
| AA-Omniscience | 0.3 | -17.1 |
| Artificial Analysis Coding Index | 48.7 | 14.5 |
| critpt | 11.6 | 0 |
| Fortress | 17.5 | 20.6 |
| gdpval | 48.3 | 0 |
| GPQA Diamond | 92.4 | 51.5 |
| HLE | 45.5 | 4.2 |
| IFBench | 75.4 | 39 |
| LiveCodeBench | 89 | 30.1 |
| longbench_v2 | 54.5 | 36.1 |
| MMLU-Pro | 87 | 73.4 |
| mmmlu | 89.6 | 73.8 |
| OmniScience Accuracy | 44.3 | 23.2 |
| OmniScience Non-Hallucination | 38.4 | 49 |
| scicode | 52.1 | 29.9 |
| simplebench | 45.8 | 23 |
| SWE-bench Pro | 55.6 | 11.2 |
| SWE-bench Verified | 80 | 24.5 |
| Terminal-Bench Hard | 62.2 | 6.8 |
| τ²-Bench Telecom (AA run) | 98.7 | 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.