| Benchmark | DeepSeek-V4-Flash | o1 |
|---|---|---|
| AA Agentic Index | 61.3 | 31.1 |
| AA Intelligence | 51.8 | 24 |
| AA-LCR | 74.3 | 63.3 |
| AA-Omniscience | -14.3 | -10.6 |
| Artificial Analysis Coding Index | 69.1 | 39.7 |
| Codeforces | 3052 | 2061 |
| critpt | 16.6 | 0.3 |
| gdpval | 52.9 | 11.5 |
| GPQA Diamond | 90.8 | 75.7 |
| HLE | 38.6 | 7.7 |
| IFBench | 79.2 | 70.3 |
| OmniScience Accuracy | 40.4 | 34.7 |
| OmniScience Non-Hallucination | 15.6 | 30.7 |
| scicode | 49.9 | 35.8 |
| simplebench | 46.3 | 41.7 |
| strongreject | 97.4 | 97 |
| SWE-bench Verified | 79 | 48.9 |
| Terminal-Bench Hard | 38.6 | 12.9 |
| τ²-Bench Telecom (AA run) | 95.6 | 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.