| Benchmark | GPT-5.6 Sol | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 61 | 10 |
| AA-LCR | 77.7 | 20.3 |
| AA-Omniscience | 22 | -52.2 |
| Artificial Analysis Coding Index | 78.3 | 11.9 |
| critpt | 32.3 | 0 |
| GPQA Diamond | 94.6 | 49.1 |
| HLE | 49.5 | 4.2 |
| IFBench | 72.7 | 36.9 |
| longbench_v2 | 67.1 | 39.4 |
| OmniScience Accuracy | 59.4 | 17.6 |
| OmniScience Non-Hallucination | 10.6 | 15.3 |
| scicode | 56.9 | 26.7 |
| Terminal-Bench Hard | 65.9 | 4.5 |
| τ²-Bench Telecom (AA run) | 85.1 | 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.