| Benchmark | DeepSeek-V3 | GPT-5.6 Sol |
|---|---|---|
| AA Agentic Index | 1.6 | 57.8 |
| AA Intelligence | 15.4 | 61 |
| AA-LCR | 41.3 | 77.7 |
| AA-Omniscience | -37.6 | 22 |
| Artificial Analysis Coding Index | 23 | 78.3 |
| critpt | 0 | 32.3 |
| gdpval | 0 | 61.1 |
| GPQA Diamond | 68.4 | 94.6 |
| HLE | 5.2 | 49.5 |
| IFBench | 41 | 72.7 |
| longbench_v2 | 48.7 | 67.1 |
| OmniScience Accuracy | 25.4 | 59.4 |
| OmniScience Non-Hallucination | 23.3 | 10.6 |
| scicode | 35.8 | 56.9 |
| simplebench | 27.2 | 64.8 |
| TauBench V3 - Banking | 4.7 | 44.3 |
| Terminal-Bench 2.1 | 16.9 | 89.5 |
| Terminal-Bench Hard | 15.2 | 65.9 |
| τ²-Bench Telecom (AA run) | 47.1 | 85.1 |
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.