| Benchmark | DeepSeek-V3 | Opus 5 |
|---|---|---|
| AA Agentic Index | 1.6 | 59.2 |
| AA Intelligence | 15.4 | 63.1 |
| AA-LCR | 41.3 | 78.7 |
| AA-Omniscience | -37.6 | 37.1 |
| Artificial Analysis Coding Index | 23 | 78 |
| critpt | 0 | 29.1 |
| gdpval | 0 | 67.2 |
| GPQA Diamond | 68.4 | 93.7 |
| HLE | 5.2 | 64.7 |
| OmniScience Accuracy | 25.4 | 60.9 |
| OmniScience Non-Hallucination | 23.3 | 40.5 |
| scicode | 35.8 | 55.7 |
| simplebench | 27.2 | 80.6 |
| SWE-bench Verified | 42 | 96 |
| TauBench V3 - Banking | 4.7 | 44.7 |
| Terminal-Bench 2.1 | 16.9 | 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.