| Benchmark | DeepSeek-V4-Flash | GPT-4o |
|---|---|---|
| AA Agentic Index | 61.3 | 9.7 |
| AA Intelligence | 51.8 | 11.2 |
| AA-LCR | 74.3 | 55 |
| AA-Omniscience | -14.3 | -10.5 |
| arc_agi_1 | 84 | 4.5 |
| ARC-AGI-2 | 46 | 0 |
| Artificial Analysis Coding Index | 69.1 | 24.2 |
| Chinese SimpleQA (C-SimpleQA) | 78.9 | 64.6 |
| Codeforces | 3052 | 759 |
| critpt | 16.6 | 0 |
| gdpval | 52.9 | 0 |
| GPQA Diamond | 90.8 | 70.1 |
| HLE | 38.6 | 5.3 |
| IFBench | 79.2 | 36 |
| LiveCodeBench | 91.6 | 38.3 |
| MMLU-Pro | 86.4 | 74.7 |
| OmniScience Accuracy | 40.4 | 23.7 |
| OmniScience Non-Hallucination | 15.6 | 62.1 |
| scicode | 49.9 | 33.4 |
| simplebench | 46.3 | 17.8 |
| SWE-bench Verified | 79 | 38.8 |
| Terminal-Bench Hard | 38.6 | 8.3 |
| τ²-Bench Telecom (AA run) | 95.6 | 28.9 |
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.