| Benchmark | Claude Sonnet 4.6 | DeepSeek-V4-Flash |
|---|---|---|
| AA Agentic Index | 61.6 | 61.3 |
| AA Intelligence | 48.4 | 51.8 |
| AA-LCR | 74 | 74.3 |
| AA-Omniscience | 12.2 | -14.3 |
| arc_agi_1 | 86.5 | 84 |
| ARC-AGI-2 | 60.4 | 46 |
| arena_elo | 1472 | 1439 |
| arena_text_factuality | 1460 | 1434 |
| Artificial Analysis Coding Index | 63 | 69.1 |
| browsecomp | 76.2 | 73.2 |
| coding_arena_elo | 1523 | 1431 |
| critpt | 3.1 | 16.6 |
| gdpval | 54.8 | 52.9 |
| GDPval-AA (Elo) | 1676 | 1395 |
| GDPval-AA v2 | 1381 | 1189 |
| GPQA Diamond | 89.9 | 90.8 |
| HLE | 49 | 38.6 |
| HLE (with tools) | 46.8 | 45.1 |
| IFBench | 56.6 | 79.2 |
| itbenchSre | 39.8 | 31.5 |
| livebench_agentic_coding | 42.6 | 46.8 |
| livebench_coding | 79.3 | 75 |
| livebench_data_analysis | 78 | 79.3 |
| livebench_instruction_following | 63.2 | 65.5 |
| livebench_language | 76.1 | 79.2 |
| livebench_math | 87 | 86.8 |
| livebench_reasoning | 84.8 | 86.6 |
| MCP Atlas | 69.5 | 69 |
| OmniScience Accuracy | 40.9 | 40.4 |
| OmniScience Non-Hallucination | 51.6 | 15.6 |
| scicode | 47 | 49.9 |
| simpleqa_verified | 29 | 45 |
| SWE-bench Pro | 58.1 | 52.6 |
| SWE-bench Verified | 79.6 | 79 |
| TauBench V3 - Banking | 34.4 | 39.4 |
| Terminal-Bench 2.0 | 59.1 | 56.9 |
| Terminal-Bench 2.1 | 71.2 | 82.7 |
| Terminal-Bench Hard | 59.1 | 38.6 |
| toolathlon | 49.4 | 70.3 |
| τ²-Bench Telecom (AA run) | 97.9 | 95.6 |
| τ³-Bench | 30.5 | 22.9 |
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.