| Benchmark | DeepSeek-V3.2 | Opus 5 |
|---|---|---|
| AA Agentic Index | 39.8 | 59.2 |
| AA Intelligence | 32.8 | 63.1 |
| AA-LCR | 70.7 | 78.7 |
| AA-Omniscience | -22.5 | 37.1 |
| arc_agi_1 | 57 | 97.5 |
| ARC-AGI-2 | 4 | 90.4 |
| arena_text_factuality | 1433 | 1481 |
| Artificial Analysis Coding Index | 44.2 | 78 |
| browsecomp | 51.4 | 90.8 |
| coding_arena_elo | 1368 | 1662 |
| critpt | 2.9 | 29.1 |
| gdpval | 18.8 | 67.2 |
| GPQA Diamond | 84 | 93.7 |
| HealthBench | 46.9 | 67.1 |
| HLE | 40.8 | 64.7 |
| HLE (with tools) | 40.8 | 64.7 |
| MCP Atlas | 62.2 | 85.8 |
| OmniScience Accuracy | 33 | 60.9 |
| OmniScience Non-Hallucination | 17.3 | 40.5 |
| scicode | 39 | 55.7 |
| simpleqa_verified | 27.5 | 56.7 |
| SWE-bench Multilingual | 70.2 | 89.5 |
| SWE-bench Pro | 15.6 | 79.2 |
| SWE-bench Verified | 73.1 | 96 |
| TauBench V3 - Banking | 18.8 | 44.7 |
| Terminal-Bench 2.1 | 46.8 | 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.