GLM-4.5: benchmark results
Zhipu's open GLM-4.5 MoE model for reasoning, coding, and agentic tasks (355B total, 32B active). Provider: Zhipu. Released 2025-07-28. Access: Open.
Aggregate Rating 1668 ± 7, rank #275 of 1656 rated models, from 416 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GMP - Co-occurring Violations | 58 | Macro-F1 (x100) over the 12 violation categories of GMP Task | 100 |
| LLM Stats (TAU-bench Retail; GLM optimized user simulator) | 79.7 | Score (%) | 100 |
| UGI Leaderboard | 58.64 | UGI Score | 98.3 |
| ReLE - Agents - BFCL-V3 | 75.5 | Accuracy (%) | 97.5 |
| ReLE - Agents and Tool Use | 72.7 | Accuracy (%) | 97 |
| OpenCompass Reasoning - Puzzle - English (CompassBench 2507) | 90.8 | Score (%) | 94.7 |
| OpenCompass Knowledge - Science - English (CompassBench 2507) | 92 | Score (%) | 94.4 |
| OpenCompass Language - NLP - English (CompassBench 2507) | 96.9 | Score (%) | 94.4 |
| OpenCompass Math - Reasoning Computation - English (CompassBench 2507) | 82.5 | Score (%) | 94.4 |
| BenchTable - Utility | 77.7 | Weighted Score (%) | 93.8 |
| WildToolBench | 12.11 | Session accuracy (%): share of the 256 dialogues in which al | 93.8 |
| ZeroEval MATH-500 | 98.2 | MATH-500 Score | 93.8 |
Interactive version: theaggregate.ai/model?slug=glm-4-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.