| Benchmark | GPT-5.4 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 53.1 | 10 |
| AA-LCR | 77.7 | 20.3 |
| AA-Omniscience | 5.8 | -52.2 |
| Artificial Analysis Coding Index | 71.1 | 11.9 |
| Chinese SimpleQA (C-SimpleQA) | 76.8 | 52.2 |
| Codeforces | 3168 | 24.8 |
| critpt | 23.4 | 0 |
| GPQA Diamond | 93 | 49.1 |
| HLE | 43.7 | 4.2 |
| IFBench | 73.9 | 36.9 |
| livebench | 80.3 | 52.3 |
| MMLU-Pro | 87.5 | 71.6 |
| OmniScience Accuracy | 50.9 | 17.6 |
| OmniScience Non-Hallucination | 17.4 | 15.3 |
| scicode | 56.6 | 26.7 |
| Terminal-Bench Hard | 57.6 | 4.5 |
| τ²-Bench Telecom (AA run) | 87.1 | 34.5 |
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.