| Benchmark | GLM-5.2 Full Open Source | GPT-5 |
|---|---|---|
| AA Agentic Index | 45.7 | 49.7 |
| AA Intelligence | 52.6 | 35.3 |
| AA-LCR | 76.7 | 76.3 |
| AA-Omniscience | 4.4 | -8.7 |
| arc_agi_1 | 77 | 56.2 |
| ARC-AGI-2 | 22.8 | 7.5 |
| Artificial Analysis Coding Index | 68.8 | 38.9 |
| coding_arena_elo | 1582 | 1419 |
| critpt | 20.9 | 5.7 |
| gdpval | 50.3 | 32.5 |
| GPQA Diamond | 91.2 | 87.3 |
| HLE | 54.7 | 28.5 |
| HLE (with tools) | 54.7 | 35.2 |
| HMMT 2025 | 94.4 | 93.3 |
| hmmt_nov_2025 | 94.4 | 89.2 |
| IFBench | 73.3 | 73.1 |
| imo_answer_bench | 91 | 76 |
| OmniScience Accuracy | 24.3 | 40.3 |
| OmniScience Non-Hallucination | 73.7 | 21 |
| scicode | 50.5 | 43 |
| simplebench | 58.8 | 56.7 |
| simpleqa_verified | 38.1 | 50.6 |
| SWE-bench Pro | 62.1 | 41.8 |
| TauBench V3 - Banking | 34.6 | 22.1 |
| Terminal-Bench 2.1 | 82.7 | 35.2 |
| Terminal-Bench Hard | 50.8 | 37.9 |
| τ²-Bench Telecom (AA run) | 99.1 | 86.5 |
| τ³-Bench | 26.8 | 19.6 |
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.