| Benchmark | DeepSeek-V3.2 | GPT-4o |
|---|---|---|
| AA Agentic Index | 39.8 | 9.7 |
| AA Intelligence | 32.8 | 11.2 |
| AA-LCR | 70.7 | 55 |
| AA-Omniscience | -22.5 | -10.5 |
| AIME 2025 | 93.1 | 11.7 |
| arc_agi_1 | 57 | 4.5 |
| ARC-AGI-2 | 4 | 0 |
| Artificial Analysis Coding Index | 44.2 | 24.2 |
| critpt | 2.9 | 0 |
| gaia_level1 | 90.3 | 39.8 |
| gaia_level2 | 73.6 | 27 |
| gaia_level3 | 61.2 | 14.3 |
| gaia_overall | 76.7 | 28.9 |
| gdpval | 18.8 | 0 |
| GPQA Diamond | 84 | 70.1 |
| HLE | 40.8 | 5.3 |
| IFBench | 61 | 36 |
| LiveCodeBench | 86 | 38.3 |
| LiveCodeBench v6 | 83.3 | 30.9 |
| longbench_v2 | 59.8 | 48.1 |
| mmlu_redux | 93.7 | 88 |
| MMLU-Pro | 86 | 74.7 |
| OmniScience Accuracy | 33 | 23.7 |
| OmniScience Non-Hallucination | 17.3 | 62.1 |
| scicode | 39 | 33.4 |
| swe_bench_bash | 60 | 21.6 |
| SWE-bench Verified | 73.1 | 38.8 |
| Terminal-Bench Hard | 35.6 | 8.3 |
| vectara_answer_rate | 92.6 | 93.8 |
| vectara_avg_summary_length | 62 | 86.6 |
| vectara_factual_consistency | 93.7 | 90.4 |
| vectara_hallucination_rate ↓ | 6.3 | 9.6 |
| τ²-Bench Telecom (AA run) | 90.6 | 28.9 |
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.