| Benchmark | DeepSeek-V3 | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 1.6 | 45.7 |
| AA Intelligence | 15.4 | 52.6 |
| AA-LCR | 41.3 | 76.7 |
| AA-Omniscience | -37.6 | 4.4 |
| Artificial Analysis Coding Index | 23 | 68.8 |
| critpt | 0 | 20.9 |
| gdpval | 0 | 50.3 |
| GPQA Diamond | 68.4 | 91.2 |
| HLE | 5.2 | 54.7 |
| HMMT 2025 | 27.5 | 94.4 |
| IFBench | 41 | 73.3 |
| OmniScience Accuracy | 25.4 | 24.3 |
| OmniScience Non-Hallucination | 23.3 | 73.7 |
| scicode | 35.8 | 50.5 |
| simplebench | 27.2 | 58.8 |
| TauBench V3 - Banking | 4.7 | 34.6 |
| Terminal-Bench 2.1 | 16.9 | 82.7 |
| Terminal-Bench Hard | 15.2 | 50.8 |
| τ²-Bench Telecom (AA run) | 47.1 | 99.1 |
| τ³-Bench | 4.7 | 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.