GLM-5.2: benchmark results
Zhipu's flagship GLM-5.2 model for reasoning, coding, and agentic tasks. Provider: Zhipu. Released 2026-06-13. Access: Open.
Aggregate Rating 1789 ± 3, rank #73 of 1656 rated models, from 1253 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| APIFlow-Bench 1.0 - 15-Subtask Chain | 74.55 | Pass Rate (%) | 100 |
| ArkEval - Compile@1 | 54.38 | Compile@1 (%): share of instances whose patch completes the | 100 |
| CESBench | 83.6 | Composite score (%; unweighted mean of the four task-type me | 100 |
| CESBench - Fault Injection | 81.7 | Composite score (%; mean over the four task types of the ite | 100 |
| CESBench - Implementation | 85.4 | Composite score (%; mean over the four task types of the ite | 100 |
| CESBench - Integration | 90.7 | Composite score (%; mean over the four task types of the ite | 100 |
| CESBench - Judgment | 58.8 | Judgment score (%; the 67 true-or-false security claims: a w | 100 |
| EntLORE - Explicit Lookup (Gold Documents) | 89.2 | Answer accuracy (%; L1, 469 questions answered by a stated f | 100 |
| EntLORE - Explicit Lookup (GraphRAG) | 68.5 | Answer accuracy (%; L1, 469 questions answered by a stated f | 100 |
| EntLORE - Explicit Lookup (LLM Wiki) | 67 | Answer accuracy (%; L1, 469 questions answered by a stated f | 100 |
| EntLORE - Latent Organizational Reasoning (GraphRAG) | 43.1 | Answer accuracy (%; L3, 234 questions whose target relation | 100 |
| Guesswork 2026-07 | 0.67 | MAE (z-score units) | 100 |
Interactive version: theaggregate.ai/model?slug=glm-5-2 · How It Works · Data refreshed daily, snapshot 2026-10-20.