| Benchmark | DeepSeek-R1 | DeepSeek-V3.2 |
|---|---|---|
| AA Agentic Index | 3.1 | 39.8 |
| AA Intelligence | 18.6 | 32.8 |
| AA-LCR | 56 | 70.7 |
| AA-Omniscience | -31.3 | -22.5 |
| AIME 2025 | 87.5 | 93.1 |
| arc_agi_1 | 21.2 | 57 |
| ARC-AGI-2 | 1.3 | 4 |
| Artificial Analysis Coding Index | 24.6 | 44.2 |
| browsecomp | 8.9 | 51.4 |
| browsecomp_zh | 35.7 | 65 |
| critpt | 0.6 | 2.9 |
| gdpval | 1.5 | 18.8 |
| GPQA Diamond | 81 | 84 |
| HLE | 17.7 | 40.8 |
| HMMT 2025 | 79.4 | 90.2 |
| hmmt_feb_2025 | 76.7 | 92.5 |
| IFBench | 39 | 61 |
| LiveCodeBench | 84.4 | 86 |
| longbench_v2 | 58.3 | 59.8 |
| mmlu_redux | 93.4 | 93.7 |
| MMLU-Pro | 85 | 86 |
| OmniScience Accuracy | 30.7 | 33 |
| OmniScience Non-Hallucination | 10.5 | 17.3 |
| scicode | 35.7 | 39 |
| simpleqa_verified | 27.4 | 27.5 |
| SWE-bench Multilingual | 30.5 | 70.2 |
| SWE-bench Verified | 57.6 | 73.1 |
| TauBench V3 - Banking | 6.4 | 18.8 |
| Terminal-Bench | 5.7 | 37.7 |
| Terminal-Bench 2.1 | 19.1 | 46.8 |
| Terminal-Bench Hard | 6.1 | 35.6 |
| vectara_answer_rate | 97 | 92.6 |
| vectara_avg_summary_length | 93.5 | 62 |
| vectara_factual_consistency | 88.7 | 93.7 |
| vectara_hallucination_rate ↓ | 11.3 | 6.3 |
| τ²-Bench Telecom (AA run) | 11.4 | 90.6 |
| τ³-Bench | 6.4 | 69.2 |
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.