| Benchmark | DeepSeek-V3 | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 1.6 | 58.2 |
| AA Intelligence | 15.4 | 53.1 |
| AA-LCR | 41.3 | 77.7 |
| AA-Omniscience | -37.6 | 5.8 |
| Artificial Analysis Coding Index | 23 | 71.1 |
| Chinese SimpleQA (C-SimpleQA) | 68 | 76.8 |
| Codeforces | 1134 | 3168 |
| critpt | 0 | 23.4 |
| gdpval | 0 | 50.1 |
| GPQA Diamond | 68.4 | 93 |
| HLE | 5.2 | 43.7 |
| IFBench | 41 | 73.9 |
| livebench | 72.4 | 80.3 |
| MMLU-Pro | 81.2 | 87.5 |
| OmniScience Accuracy | 25.4 | 50.9 |
| OmniScience Non-Hallucination | 23.3 | 17.4 |
| scicode | 35.8 | 56.6 |
| TauBench V3 - Banking | 4.7 | 39.6 |
| Terminal-Bench 2.1 | 16.9 | 78.3 |
| Terminal-Bench Hard | 15.2 | 57.6 |
| vectara_answer_rate | 97.5 | 99.9 |
| vectara_avg_summary_length | 81.7 | 81.7 |
| vectara_factual_consistency | 93.9 | 93 |
| vectara_hallucination_rate ↓ | 6.1 | 7 |
| τ²-Bench Telecom (AA run) | 47.1 | 87.1 |
| τ³-Bench | 4.7 | 72.9 |
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.