| Benchmark | Claude 3.7 Sonnet | DeepSeek-V3.2 |
|---|---|---|
| AA Agentic Index | 37 | 39.8 |
| AA Intelligence | 27.6 | 32.8 |
| AA-LCR | 62.3 | 70.7 |
| AA-Omniscience | -0.7 | -22.5 |
| AIME 2025 | 54.8 | 93.1 |
| arc_agi_1 | 13.6 | 57 |
| ARC-AGI-2 | 0 | 4 |
| Artificial Analysis Coding Index | 36.4 | 44.2 |
| critpt | 0.9 | 2.9 |
| gdpval | 27.4 | 18.8 |
| GPQA Diamond | 84.8 | 84 |
| HLE | 9.7 | 40.8 |
| hmmt_feb_2025 | 31.7 | 92.5 |
| IFBench | 48.3 | 61 |
| OmniScience Accuracy | 28.2 | 33 |
| OmniScience Non-Hallucination | 60 | 17.3 |
| scicode | 40.3 | 39 |
| swe_bench_bash | 52.8 | 60 |
| SWE-bench Verified | 70.3 | 73.1 |
| Terminal-Bench | 35.2 | 37.7 |
| Terminal-Bench Hard | 21.2 | 35.6 |
| τ²-Bench Telecom (AA run) | 54.7 | 90.6 |
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.