| Benchmark | DeepSeek-R1 | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 3.1 | 60.2 |
| AA Intelligence | 18.6 | 43.3 |
| AA-LCR | 56 | 79.3 |
| AA-Omniscience | -31.3 | 0.3 |
| AIME 2025 | 87.5 | 100 |
| arc_agi_1 | 21.2 | 86.2 |
| ARC-AGI-2 | 1.3 | 52.9 |
| Artificial Analysis Coding Index | 24.6 | 48.7 |
| browsecomp | 8.9 | 65.8 |
| browsecomp_zh | 35.7 | 76.1 |
| critpt | 0.6 | 11.6 |
| Fortress | 74.4 | 17.5 |
| gdpval | 1.5 | 48.3 |
| GPQA Diamond | 81 | 92.4 |
| HLE | 17.7 | 45.5 |
| HMMT 2025 | 79.4 | 99.4 |
| hmmt_feb_2025 | 76.7 | 100 |
| IFBench | 39 | 75.4 |
| LiveCodeBench | 84.4 | 89 |
| longbench_v2 | 58.3 | 54.5 |
| MMLU-Pro | 85 | 87 |
| MultiNRC | 27.6 | 42.2 |
| OmniScience Accuracy | 30.7 | 44.3 |
| OmniScience Non-Hallucination | 10.5 | 38.4 |
| scicode | 35.7 | 52.1 |
| simplebench | 40.8 | 45.8 |
| simpleqa_verified | 27.4 | 38.9 |
| SWE-bench Multilingual | 30.5 | 72 |
| SWE-bench Verified | 57.6 | 80 |
| TauBench V3 - Banking | 6.4 | 11.1 |
| Terminal-Bench Hard | 6.1 | 62.2 |
| vectara_answer_rate | 97 | 100 |
| vectara_avg_summary_length | 93.5 | 186.3 |
| vectara_factual_consistency | 88.7 | 89.2 |
| vectara_hallucination_rate ↓ | 11.3 | 10.8 |
| τ²-Bench Telecom (AA run) | 11.4 | 98.7 |
| τ³-Bench | 6.4 | 12 |
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.