| Benchmark | DeepSeek-R1 | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 3.1 | 58.2 |
| AA Intelligence | 18.6 | 53.1 |
| AA-LCR | 56 | 77.7 |
| AA-Omniscience | -31.3 | 5.8 |
| arc_agi_1 | 21.2 | 93.7 |
| ARC-AGI-2 | 1.3 | 74 |
| Artificial Analysis Coding Index | 24.6 | 71.1 |
| browsecomp | 8.9 | 82.7 |
| Chinese SimpleQA (C-SimpleQA) | 63.7 | 76.8 |
| Codeforces | 2029 | 3168 |
| critpt | 0.6 | 23.4 |
| gdpval | 1.5 | 50.1 |
| GPQA Diamond | 81 | 93 |
| HLE | 17.7 | 43.7 |
| IFBench | 39 | 73.9 |
| MMLU-Pro | 85 | 87.5 |
| MultiNRC | 27.6 | 58.3 |
| OmniScience Accuracy | 30.7 | 50.9 |
| OmniScience Non-Hallucination | 10.5 | 17.4 |
| scicode | 35.7 | 56.6 |
| simpleqa_verified | 27.4 | 45.3 |
| SWE-bench Multilingual | 30.5 | 71.7 |
| TauBench V3 - Banking | 6.4 | 39.6 |
| Terminal-Bench 2.1 | 19.1 | 78.3 |
| Terminal-Bench Hard | 6.1 | 57.6 |
| vectara_answer_rate | 97 | 99.9 |
| vectara_avg_summary_length | 93.5 | 81.7 |
| vectara_factual_consistency | 88.7 | 93 |
| vectara_hallucination_rate ↓ | 11.3 | 7 |
| τ²-Bench Telecom (AA run) | 11.4 | 87.1 |
| τ³-Bench | 6.4 | 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.