GLM-4.5 (Reasoning): benchmark results

GLM-4.5 evaluated with reasoning enabled. Provider: Zhipu. Released 2025-07-28. Access: Open.

Unified ELO 1564 ± 1, rank #584 of 1761 rated models, from 40 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UGI - Writing57.34Writing Score91.7
UGI - Natural Intelligence46.56NatInt Score88.9
AA Omniscience - Software Engineering (SWE) - Rust62Accuracy (%)84
AA Omniscience - Software Engineering (SWE) - Go30.61Accuracy (%)83.1
AA MATH-50097.87Accuracy (%)82
AA MMLU-Pro83.51Accuracy (%)81.8
AA Omniscience - Software Engineering (SWE) - C49.49Accuracy (%)80.5
AA Omniscience - Software Engineering (SWE) - Kotlin28Accuracy (%)80.5
AA Omniscience - Software Engineering (SWE) - Julia24Accuracy (%)80.2
AA LiveCodeBench73.76Pass@1 (%)80
AA Omniscience - Software Engineering (SWE) - R20Accuracy (%)79.5
AA Omniscience - Software Engineering (SWE) - TypeScript33.33Accuracy (%)79.4

Interactive version: theaggregate.ai/model?slug=glm-4-5-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-05.