| Benchmark | DeepSeek-V4-Flash | o3 |
|---|---|---|
| AA Agentic Index | 61.3 | 36.1 |
| AA Intelligence | 51.8 | 31.1 |
| AA-LCR | 74.3 | 73.3 |
| AA-Omniscience | -14.3 | -15.3 |
| arc_agi_1 | 84 | 60.8 |
| ARC-AGI-2 | 46 | 6.5 |
| Artificial Analysis Coding Index | 69.1 | 38.4 |
| browsecomp | 73.2 | 49.7 |
| critpt | 16.6 | 1.1 |
| gdpval | 52.9 | 12.8 |
| GPQA Diamond | 90.8 | 83.3 |
| HLE | 38.6 | 20.3 |
| IFBench | 79.2 | 71.4 |
| LiveCodeBench | 91.6 | 84.7 |
| MMLU-Pro | 86.4 | 85 |
| OmniScience Accuracy | 40.4 | 38.6 |
| OmniScience Non-Hallucination | 15.6 | 12.9 |
| scicode | 49.9 | 41 |
| simplebench | 46.3 | 53.1 |
| simpleqa_verified | 45 | 53 |
| strongreject | 97.4 | 97 |
| SWE-bench Verified | 79 | 69.1 |
| Terminal-Bench Hard | 38.6 | 37.1 |
| τ²-Bench Telecom (AA run) | 95.6 | 80.7 |
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.