| Benchmark | Claude Sonnet 4.5 | DeepSeek-R1 |
|---|---|---|
| AA Agentic Index | 50.6 | 3.1 |
| AA Intelligence | 37.4 | 18.6 |
| AA-LCR | 68.3 | 56 |
| AA-Omniscience | -0.1 | -31.3 |
| AIME 2025 | 88 | 87.5 |
| AIR-Bench 2024 | 89.8 | 52.9 |
| arc_agi_1 | 63.7 | 21.2 |
| ARC-AGI-2 | 13.6 | 1.3 |
| Artificial Analysis Coding Index | 52.1 | 24.6 |
| browsecomp | 24.1 | 8.9 |
| browsecomp_zh | 42.4 | 35.7 |
| critpt | 1.1 | 0.6 |
| Fortress | 17 | 74.4 |
| gdpval | 40.4 | 1.5 |
| GPQA Diamond | 83.4 | 81 |
| HLE | 19.8 | 17.7 |
| HMMT 2025 | 74.6 | 79.4 |
| hmmt_feb_2025 | 79.2 | 76.7 |
| IFBench | 57.3 | 39 |
| LiveCodeBench | 71 | 84.4 |
| longbench_v2 | 61.8 | 58.3 |
| mmlu_redux | 95.6 | 93.4 |
| MMLU-Pro | 88.2 | 85 |
| multichallenge | 55.3 | 45 |
| MultiNRC | 35.8 | 27.6 |
| OmniScience Accuracy | 32.9 | 30.7 |
| OmniScience Non-Hallucination | 50.9 | 10.5 |
| scicode | 45 | 35.7 |
| simplebench | 54.3 | 40.8 |
| simpleqa_verified | 23.6 | 27.4 |
| SWE-bench Multilingual | 68 | 30.5 |
| SWE-bench Verified | 82 | 57.6 |
| TAU-bench (airline) | 70 | 53.5 |
| TAU-bench (retail) | 86.2 | 63.9 |
| TauBench V3 - Banking | 24.5 | 6.4 |
| Terminal-Bench | 51 | 5.7 |
| Terminal-Bench 2.1 | 55.8 | 19.1 |
| Terminal-Bench Hard | 35.6 | 6.1 |
| vectara_answer_rate | 95.6 | 97 |
| vectara_avg_summary_length | 127.8 | 93.5 |
| vectara_factual_consistency | 88 | 88.7 |
| vectara_hallucination_rate ↓ | 12 | 11.3 |
| τ²-Bench Telecom (AA run) | 78.1 | 11.4 |
| τ³-Bench | 19 | 6.4 |
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.