| Benchmark | DeepSeek-V3.2 | o1 |
|---|---|---|
| AA Agentic Index | 39.8 | 31.1 |
| AA Intelligence | 32.8 | 24 |
| AA-LCR | 70.7 | 63.3 |
| AA-Omniscience | -22.5 | -10.6 |
| AIME 2025 | 93.1 | 81.7 |
| Artificial Analysis Coding Index | 44.2 | 39.7 |
| critpt | 2.9 | 0.3 |
| gdpval | 18.8 | 11.5 |
| GPQA Diamond | 84 | 75.7 |
| HLE | 40.8 | 7.7 |
| IFBench | 61 | 70.3 |
| OmniScience Accuracy | 33 | 34.7 |
| OmniScience Non-Hallucination | 17.3 | 30.7 |
| scicode | 39 | 35.8 |
| SWE-bench Verified | 73.1 | 48.9 |
| Terminal-Bench Hard | 35.6 | 12.9 |
| τ²-Bench Telecom (AA run) | 90.6 | 62.6 |
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.