| Benchmark | GPT-5.4 | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 58.2 | 6.3 |
| AA Intelligence | 53.1 | 8.5 |
| AA-LCR | 77.7 | 24.7 |
| AA-Omniscience | 5.8 | -17.1 |
| Artificial Analysis Coding Index | 71.1 | 14.5 |
| Chinese SimpleQA (C-SimpleQA) | 76.8 | 54.7 |
| Codeforces | 3168 | 25.3 |
| critpt | 23.4 | 0 |
| gdpval | 50.1 | 0 |
| GPQA Diamond | 93 | 51.5 |
| HLE | 43.7 | 4.2 |
| IFBench | 73.9 | 39 |
| MMLU-Pro | 87.5 | 73.4 |
| OmniScience Accuracy | 50.9 | 23.2 |
| OmniScience Non-Hallucination | 17.4 | 49 |
| scicode | 56.6 | 29.9 |
| SWE-bench Pro | 59.1 | 11.2 |
| Terminal-Bench Hard | 57.6 | 6.8 |
| τ²-Bench Telecom (AA run) | 87.1 | 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.