| Benchmark | Claude Opus 4.5 | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 59.6 | 1.6 |
| AA Intelligence | 41.9 | 15.4 |
| AA-LCR | 76 | 41.3 |
| AA-Omniscience | 14 | -37.6 |
| AIME 2025 | 92.8 | 51.3 |
| Artificial Analysis Coding Index | 47.8 | 23 |
| C-Eval | 92.2 | 90.1 |
| critpt | 4.6 | 0 |
| gdpval | 47.3 | 0 |
| GPQA Diamond | 87 | 68.4 |
| HLE | 30.8 | 5.2 |
| hmmt_feb_2025 | 92.9 | 29.2 |
| IFBench | 58 | 41 |
| livebench | 76 | 72.4 |
| LiveCodeBench | 87 | 49.6 |
| LiveCodeBench v6 | 84.8 | 46.9 |
| longbench_v2 | 64.4 | 48.7 |
| mmlu_redux | 95.6 | 90.5 |
| MMLU-Pro | 90 | 81.2 |
| mmmlu | 90.8 | 79.4 |
| multichallenge | 59 | 31.4 |
| OmniScience Accuracy | 46.6 | 25.4 |
| OmniScience Non-Hallucination | 39 | 23.3 |
| scicode | 50 | 35.8 |
| simplebench | 62 | 27.2 |
| supergpqa | 70.6 | 53.7 |
| SWE-bench Verified | 81.5 | 42 |
| Tau2 retail | 88.9 | 69.1 |
| Terminal-Bench Hard | 47 | 15.2 |
| vectara_answer_rate | 98.7 | 97.5 |
| vectara_avg_summary_length | 114.5 | 81.7 |
| vectara_factual_consistency | 89.1 | 93.9 |
| vectara_hallucination_rate ↓ | 10.9 | 6.1 |
| τ²-Bench | 98.2 | 32.5 |
| τ²-Bench Telecom (AA run) | 89.5 | 47.1 |
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.