| Benchmark | Claude Sonnet 4 | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 39.2 | 45.7 |
| AA Intelligence | 29.8 | 52.6 |
| AA-LCR | 69.7 | 76.7 |
| AA-Omniscience | 0.2 | 4.4 |
| arc_agi_1 | 63.7 | 77 |
| ARC-AGI-2 | 13.6 | 22.8 |
| Artificial Analysis Coding Index | 37.6 | 68.8 |
| critpt | 1.1 | 20.9 |
| gdpval | 31.2 | 50.3 |
| GPQA Diamond | 78 | 91.2 |
| HLE | 10.7 | 54.7 |
| HLE (with tools) | 20.3 | 54.7 |
| HMMT 2025 | 15.9 | 94.4 |
| IFBench | 55 | 73.3 |
| OmniScience Accuracy | 22.7 | 24.3 |
| OmniScience Non-Hallucination | 70.9 | 73.7 |
| scicode | 40 | 50.5 |
| simplebench | 45.5 | 58.8 |
| SWE-bench Pro | 42.7 | 62.1 |
| TauBench V3 - Banking | 16.7 | 34.6 |
| Terminal-Bench 2.1 | 36.3 | 82.7 |
| Terminal-Bench Hard | 31.1 | 50.8 |
| τ²-Bench Telecom (AA run) | 64.6 | 99.1 |
| τ³-Bench | 13.8 | 26.8 |
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.