| Benchmark | DeepSeek-V3.2 | o3 |
|---|---|---|
| AA Agentic Index | 39.8 | 36.1 |
| AA Intelligence | 32.8 | 31.1 |
| AA-LCR | 70.7 | 73.3 |
| AA-Omniscience | -22.5 | -15.3 |
| AIME 2025 | 93.1 | 89.2 |
| arc_agi_1 | 57 | 60.8 |
| ARC-AGI-2 | 4 | 6.5 |
| Artificial Analysis Coding Index | 44.2 | 38.4 |
| browsecomp | 51.4 | 49.7 |
| critpt | 2.9 | 1.1 |
| frontiermath_tier_4 | 2.1 | 2.1 |
| gdpval | 18.8 | 12.8 |
| GPQA Diamond | 84 | 83.3 |
| HLE | 40.8 | 20.3 |
| hmmt_feb_2025 | 92.5 | 77.5 |
| IFBench | 61 | 71.4 |
| LiveCodeBench | 86 | 84.7 |
| longbench_v2 | 59.8 | 58.8 |
| MMLU-Pro | 86 | 85 |
| OmniScience Accuracy | 33 | 38.6 |
| OmniScience Non-Hallucination | 17.3 | 12.9 |
| scicode | 39 | 41 |
| simpleqa_verified | 27.5 | 53 |
| swe_bench_bash | 60 | 58.4 |
| SWE-bench Verified | 73.1 | 69.1 |
| Terminal-Bench Hard | 35.6 | 37.1 |
| τ²-Bench Telecom (AA run) | 90.6 | 80.7 |
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.