| Benchmark | Claude Sonnet 4.6 | DeepSeek-V3.2 |
|---|---|---|
| AA Agentic Index | 61.6 | 39.8 |
| AA Intelligence | 48.4 | 32.8 |
| AA-LCR | 74 | 70.7 |
| AA-Omniscience | 12.2 | -22.5 |
| apexAgents | 28 | 14.5 |
| arc_agi_1 | 86.5 | 57 |
| ARC-AGI-2 | 60.4 | 4 |
| arena_text_factuality | 1460 | 1433 |
| Artificial Analysis Coding Index | 63 | 44.2 |
| browsecomp | 76.2 | 51.4 |
| coding_arena_elo | 1523 | 1368 |
| critpt | 3.1 | 2.9 |
| frontiermath_tier_4 | 8.3 | 2.1 |
| gdpval | 54.8 | 18.8 |
| GPQA (unspecified) | 89.9 | 79.9 |
| GPQA Diamond | 89.9 | 84 |
| HLE | 49 | 40.8 |
| HLE (with tools) | 46.8 | 40.8 |
| IFBench | 56.6 | 61 |
| MCP Atlas | 69.5 | 62.2 |
| OmniScience Accuracy | 40.9 | 33 |
| OmniScience Non-Hallucination | 51.6 | 17.3 |
| scicode | 47 | 39 |
| simpleqa_verified | 29 | 27.5 |
| SWE-bench Pro | 58.1 | 15.6 |
| SWE-bench Verified | 79.6 | 73.1 |
| TauBench V3 - Banking | 34.4 | 18.8 |
| Terminal-Bench 2.0 | 59.1 | 46.4 |
| Terminal-Bench 2.1 | 71.2 | 46.8 |
| Terminal-Bench Hard | 59.1 | 35.6 |
| toolathlon | 49.4 | 35.2 |
| vectara_answer_rate | 99.9 | 92.6 |
| vectara_avg_summary_length | 114.7 | 62 |
| vectara_factual_consistency | 89.4 | 93.7 |
| vectara_hallucination_rate ↓ | 10.6 | 6.3 |
| τ²-Bench Telecom (AA run) | 97.9 | 90.6 |
| τ³-Bench | 30.5 | 69.2 |
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.