| Benchmark | DeepSeek-V3 | GPT-5.5 |
|---|---|---|
| AA Agentic Index | 1.6 | 47.4 |
| AA Intelligence | 15.4 | 56.3 |
| AA-LCR | 41.3 | 79 |
| AA-Omniscience | -37.6 | 20.5 |
| Artificial Analysis Coding Index | 23 | 74.9 |
| critpt | 0 | 27.1 |
| gdpval | 0 | 49.5 |
| GPQA Diamond | 68.4 | 93.6 |
| HLE | 5.2 | 52.2 |
| IFBench | 41 | 75.9 |
| livebench | 72.4 | 80.7 |
| OmniScience Accuracy | 25.4 | 58 |
| OmniScience Non-Hallucination | 23.3 | 12 |
| scicode | 35.8 | 56.1 |
| simplebench | 27.2 | 69 |
| SWE-bench Verified | 42 | 80.6 |
| TauBench V3 - Banking | 4.7 | 39 |
| Terminal-Bench 2.1 | 16.9 | 84.3 |
| Terminal-Bench Hard | 15.2 | 60.6 |
| vectara_answer_rate | 97.5 | 100 |
| vectara_avg_summary_length | 81.7 | 129.6 |
| vectara_factual_consistency | 93.9 | 90.7 |
| vectara_hallucination_rate ↓ | 6.1 | 9.3 |
| τ²-Bench Telecom (AA run) | 47.1 | 93.9 |
| τ³-Bench | 4.7 | 31.3 |
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.