| Benchmark | DeepSeek-V3 | GPT-5.1 |
|---|---|---|
| AA Agentic Index | 1.6 | 32.2 |
| AA Intelligence | 15.4 | 37.5 |
| AA-LCR | 41.3 | 76.7 |
| AA-Omniscience | -37.6 | 5.4 |
| AIME 2025 | 51.3 | 94.2 |
| Artificial Analysis Coding Index | 23 | 49.4 |
| critpt | 0 | 4.9 |
| gdpval | 0 | 25 |
| GPQA Diamond | 68.4 | 88.1 |
| HLE | 5.2 | 28.5 |
| hmmt_feb_2025 | 29.2 | 96.3 |
| IFBench | 41 | 72.9 |
| LiveCodeBench v6 | 46.9 | 87 |
| MMLU-Pro | 81.2 | 87 |
| mmmlu | 79.4 | 91 |
| multichallenge | 31.4 | 63.4 |
| OmniScience Accuracy | 25.4 | 37.7 |
| OmniScience Non-Hallucination | 23.3 | 48.1 |
| scicode | 35.8 | 43.3 |
| simplebench | 27.2 | 53.2 |
| SWE-bench Verified | 42 | 76.3 |
| Tau2 airline | 39 | 67 |
| Tau2 retail | 69.1 | 77.9 |
| TauBench V3 - Banking | 4.7 | 15.9 |
| Terminal-Bench 2.1 | 16.9 | 52.4 |
| Terminal-Bench Hard | 15.2 | 45.5 |
| vectara_answer_rate | 97.5 | 100 |
| vectara_avg_summary_length | 81.7 | 254.4 |
| vectara_factual_consistency | 93.9 | 89.1 |
| vectara_hallucination_rate ↓ | 6.1 | 12.1 |
| τ²-Bench | 32.5 | 82.7 |
| τ²-Bench Telecom (AA run) | 47.1 | 81.9 |
| τ³-Bench | 4.7 | 14 |
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.