| Benchmark | Claude Sonnet 4.5 | DeepSeek-V4-Pro |
|---|---|---|
| AA Agentic Index | 50.6 | 63.3 |
| AA Intelligence | 37.4 | 53 |
| AA-LCR | 68.3 | 70 |
| AA-Omniscience | -0.1 | -10.6 |
| arena_elo | 1456 | 1457 |
| arena_text_factuality | 1448 | 1450 |
| Artificial Analysis Coding Index | 52.1 | 59.4 |
| browsecomp | 24.1 | 83.4 |
| coding_arena_elo | 1392 | 1582 |
| critpt | 1.1 | 13 |
| gdpval | 40.4 | 49 |
| GPQA Diamond | 83.4 | 90.5 |
| HLE | 19.8 | 48.2 |
| HLE (with tools) | 32 | 48.2 |
| hmmt_nov_2025 | 81.7 | 94.4 |
| IFBench | 57.3 | 76.5 |
| imo_answer_bench | 65.9 | 89.8 |
| LiveCodeBench | 71 | 93.5 |
| MCP Atlas | 59.5 | 74.2 |
| MMLU-Pro | 88.2 | 87.5 |
| OmniScience Accuracy | 32.9 | 43 |
| OmniScience Non-Hallucination | 50.9 | 12.2 |
| scicode | 45 | 50 |
| simplebench | 54.3 | 50.9 |
| simpleqa_verified | 23.6 | 57.9 |
| SWE-bench Multilingual | 68 | 76.2 |
| SWE-bench Pro | 43.6 | 55.4 |
| SWE-bench Verified | 82 | 80.6 |
| TauBench V3 - Banking | 24.5 | 30.1 |
| Terminal-Bench 2.0 | 50 | 67.9 |
| Terminal-Bench 2.1 | 55.8 | 72.1 |
| Terminal-Bench Hard | 35.6 | 46.2 |
| toolathlon | 41 | 51.8 |
| vectara_answer_rate | 95.6 | 97.2 |
| vectara_avg_summary_length | 127.8 | 153.8 |
| vectara_factual_consistency | 88 | 91.4 |
| vectara_hallucination_rate ↓ | 12 | 8.6 |
| τ²-Bench Telecom (AA run) | 78.1 | 96.2 |
| τ³-Bench | 19 | 25.8 |
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.