| Benchmark | Claude Opus 4.5 | GLM-5.2 Full Open Source |
|---|---|---|
| AA Agentic Index | 59.6 | 45.7 |
| AA Intelligence | 41.9 | 52.6 |
| AA-LCR | 76 | 76.7 |
| AA-Omniscience | 14 | 4.4 |
| aime_2026 | 95.1 | 99.2 |
| arc_agi_1 | 82 | 77 |
| ARC-AGI-2 | 37.6 | 22.8 |
| arena_elo | 1473 | 1471 |
| arena_text_factuality | 1458 | 1460 |
| Artificial Analysis Coding Index | 47.8 | 68.8 |
| coding_arena_elo | 1495 | 1582 |
| critpt | 4.6 | 20.9 |
| gdpval | 47.3 | 50.3 |
| GPQA Diamond | 87 | 91.2 |
| HLE | 30.8 | 54.7 |
| HLE (with tools) | 43.4 | 54.7 |
| hmmt_feb_2026 | 85.3 | 92.5 |
| hmmt_nov_2025 | 93.3 | 94.4 |
| IFBench | 58 | 73.3 |
| imo_answer_bench | 84 | 91 |
| livebench_agentic_coding | 39.7 | 51.8 |
| livebench_coding | 79.7 | 79.7 |
| livebench_data_analysis | 74.4 | 73.7 |
| livebench_instruction_following | 62.5 | 62.3 |
| livebench_language | 81.3 | 76.2 |
| livebench_math | 90.4 | 89.8 |
| livebench_reasoning | 80.1 | 78.6 |
| MCP Atlas | 69.8 | 82.6 |
| nl2repo | 43.2 | 48.9 |
| OmniScience Accuracy | 46.6 | 24.3 |
| OmniScience Non-Hallucination | 39 | 73.7 |
| scicode | 50 | 50.5 |
| simplebench | 62 | 58.8 |
| simpleqa_verified | 41.8 | 38.1 |
| SWE-bench Pro | 57.1 | 62.1 |
| Terminal-Bench Hard | 47 | 50.8 |
| tool_decathlon | 43.5 | 48.2 |
| toolathlon | 43.5 | 48.2 |
| τ²-Bench Telecom (AA run) | 89.5 | 99.1 |
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.