| Benchmark | GLM-5.2 Full Open Source | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 45.7 | 60.2 |
| AA Intelligence | 52.6 | 43.3 |
| AA-LCR | 76.7 | 79.3 |
| AA-Omniscience | 4.4 | 0.3 |
| aime_2026 | 99.2 | 98.3 |
| apexAgents | 33.7 | 23 |
| arc_agi_1 | 77 | 86.2 |
| ARC-AGI-2 | 22.8 | 52.9 |
| arena_elo | 1471 | 1437 |
| arena_text_factuality | 1460 | 1442 |
| Artificial Analysis Coding Index | 68.8 | 48.7 |
| coding_arena_elo | 1582 | 1418 |
| critpt | 20.9 | 11.6 |
| gdpval | 50.3 | 48.3 |
| GPQA Diamond | 91.2 | 92.4 |
| HLE | 54.7 | 45.5 |
| HLE (with tools) | 54.7 | 45.5 |
| HMMT 2025 | 94.4 | 99.4 |
| hmmt_feb_2026 | 92.5 | 97 |
| hmmt_nov_2025 | 94.4 | 99.2 |
| IFBench | 73.3 | 75.4 |
| imo_answer_bench | 91 | 86.3 |
| livebench_agentic_coding | 51.8 | 50.3 |
| livebench_coding | 79.7 | 76.1 |
| livebench_data_analysis | 73.7 | 78.2 |
| livebench_instruction_following | 62.3 | 61.8 |
| livebench_language | 76.2 | 79.8 |
| livebench_math | 89.8 | 93.2 |
| livebench_reasoning | 78.6 | 83.2 |
| MCP Atlas | 82.6 | 68 |
| OmniScience Accuracy | 24.3 | 44.3 |
| OmniScience Non-Hallucination | 73.7 | 38.4 |
| scicode | 50.5 | 52.1 |
| simplebench | 58.8 | 45.8 |
| simpleqa_verified | 38.1 | 38.9 |
| SWE-bench Pro | 62.1 | 55.6 |
| TauBench V3 - Banking | 34.6 | 11.1 |
| Terminal-Bench Hard | 50.8 | 62.2 |
| tool_decathlon | 48.2 | 46.3 |
| toolathlon | 48.2 | 46.3 |
| τ²-Bench Telecom (AA run) | 99.1 | 98.7 |
| τ³-Bench | 26.8 | 12 |
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.