| Benchmark | GLM-5.2 Full Open Source | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 45.7 | 6.3 |
| AA Intelligence | 52.6 | 8.5 |
| AA-LCR | 76.7 | 24.7 |
| AA-Omniscience | 4.4 | -17.1 |
| Artificial Analysis Coding Index | 68.8 | 14.5 |
| critpt | 20.9 | 0 |
| gdpval | 50.3 | 0 |
| GPQA Diamond | 91.2 | 51.5 |
| HLE | 54.7 | 4.2 |
| IFBench | 73.3 | 39 |
| OmniScience Accuracy | 24.3 | 23.2 |
| OmniScience Non-Hallucination | 73.7 | 49 |
| scicode | 50.5 | 29.9 |
| simplebench | 58.8 | 23 |
| SWE-bench Pro | 62.1 | 11.2 |
| Terminal-Bench Hard | 50.8 | 6.8 |
| τ²-Bench Telecom (AA run) | 99.1 | 19 |
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.