| Benchmark | Claude 3.7 Sonnet | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 37 | 45.7 |
| AA Intelligence | 27.6 | 52.6 |
| AA-LCR | 62.3 | 76.7 |
| AA-Omniscience | -0.7 | 4.4 |
| arc_agi_1 | 13.6 | 77 |
| ARC-AGI-2 | 0 | 22.8 |
| Artificial Analysis Coding Index | 36.4 | 68.8 |
| critpt | 0.9 | 20.9 |
| gdpval | 27.4 | 50.3 |
| GPQA Diamond | 84.8 | 91.2 |
| HLE | 9.7 | 54.7 |
| IFBench | 48.3 | 73.3 |
| OmniScience Accuracy | 28.2 | 24.3 |
| OmniScience Non-Hallucination | 60 | 73.7 |
| scicode | 40.3 | 50.5 |
| simplebench | 46.4 | 58.8 |
| Terminal-Bench Hard | 21.2 | 50.8 |
| τ²-Bench Telecom (AA run) | 54.7 | 99.1 |
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.