| Benchmark | DeepSeek-V4-Flash | Opus 5 |
|---|---|---|
| AA Agentic Index | 61.3 | 59.2 |
| AA Intelligence | 51.8 | 63.1 |
| AA-LCR | 74.3 | 78.7 |
| AA-Omniscience | -14.3 | 37.1 |
| arc_agi_1 | 84 | 97.5 |
| ARC-AGI-2 | 46 | 90.4 |
| arena_elo | 1439 | 1487 |
| arena_text_factuality | 1434 | 1481 |
| Artificial Analysis Coding Index | 69.1 | 78 |
| browsecomp | 73.2 | 90.8 |
| coding_arena_elo | 1431 | 1662 |
| critpt | 16.6 | 29.1 |
| gdpval | 52.9 | 67.2 |
| GDPval-AA (Elo) | 1395 | 1827 |
| GDPval-AA v2 | 1189 | 1861 |
| GPQA Diamond | 90.8 | 93.7 |
| HLE | 38.6 | 64.7 |
| HLE (with tools) | 45.1 | 64.7 |
| livebench_agentic_coding | 46.8 | 65.2 |
| livebench_coding | 75 | 81.5 |
| livebench_data_analysis | 79.3 | 74.5 |
| livebench_instruction_following | 65.5 | 63.8 |
| livebench_language | 79.2 | 88.7 |
| livebench_math | 86.8 | 95.7 |
| livebench_reasoning | 86.6 | 91.2 |
| MCP Atlas | 69 | 85.8 |
| OmniScience Accuracy | 40.4 | 60.9 |
| OmniScience Non-Hallucination | 15.6 | 40.5 |
| scicode | 49.9 | 55.7 |
| simplebench | 46.3 | 80.6 |
| simpleqa_verified | 45 | 56.7 |
| SWE-bench Multilingual | 73.3 | 89.5 |
| SWE-bench Pro | 52.6 | 79.2 |
| SWE-bench Verified | 79 | 96 |
| TauBench V3 - Banking | 39.4 | 44.7 |
| Terminal-Bench 2.1 | 82.7 | 89.1 |
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.