| Benchmark | Claude Sonnet 4.5 | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 50.6 | 1.6 |
| AA Intelligence | 37.4 | 15.4 |
| AA-LCR | 68.3 | 41.3 |
| AA-Omniscience | -0.1 | -37.6 |
| AIME 2025 | 88 | 51.3 |
| AIR-Bench 2024 | 89.8 | 40.8 |
| Artificial Analysis Coding Index | 52.1 | 23 |
| critpt | 1.1 | 0 |
| Frames | 85 | 73.3 |
| gdpval | 40.4 | 0 |
| GPQA Diamond | 83.4 | 68.4 |
| HLE | 19.8 | 5.2 |
| HMMT 2025 | 74.6 | 27.5 |
| hmmt_feb_2025 | 79.2 | 29.2 |
| IFBench | 57.3 | 41 |
| LiveCodeBench | 71 | 49.6 |
| LiveCodeBench v6 | 64 | 46.9 |
| longbench_v2 | 61.8 | 48.7 |
| mmlu_redux | 95.6 | 90.5 |
| MMLU-Pro | 88.2 | 81.2 |
| mmmlu | 89.1 | 79.4 |
| multichallenge | 55.3 | 31.4 |
| OmniScience Accuracy | 32.9 | 25.4 |
| OmniScience Non-Hallucination | 50.9 | 23.3 |
| scicode | 45 | 35.8 |
| simplebench | 54.3 | 27.2 |
| SWE-bench Verified | 82 | 42 |
| TauBench V3 - Banking | 24.5 | 4.7 |
| Terminal-Bench 2.1 | 55.8 | 16.9 |
| Terminal-Bench Hard | 35.6 | 15.2 |
| vectara_answer_rate | 95.6 | 97.5 |
| vectara_avg_summary_length | 127.8 | 81.7 |
| vectara_factual_consistency | 88 | 93.9 |
| vectara_hallucination_rate ↓ | 12 | 6.1 |
| τ²-Bench | 98 | 32.5 |
| τ²-Bench Telecom (AA run) | 78.1 | 47.1 |
| τ³-Bench | 19 | 4.7 |
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.