GLM-5 (Thinking): benchmark results

GLM-5 evaluated with thinking enabled. Provider: Zhipu. Released 2026-02-12. Access: Open.

Unified ELO 1645 ± 1, rank #259 of 1761 rated models, from 67 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA TAU-2 Bench98.25Accuracy (%)98.4
AA Omniscience - Software Engineering (SWE) - Dart42Accuracy (%)96.4
UGI Leaderboard54.56UGI Score95.9
AA Omniscience - Software Engineering (SWE) - Go42Accuracy (%)94
AA Omniscience - Software Engineering (SWE) - C67Accuracy (%)93.7
AA Omniscience - Software Engineering (SWE) - JavaScript52.73Accuracy (%)93.5
BenchTable74.6Total Score (%)92.7
AA Omniscience - Software Engineering (SWE) - PHP48Accuracy (%)92.3
AA Omniscience - Software Engineering (SWE) - Java31Accuracy (%)92
UGI - Natural Intelligence55.34NatInt Score92
Kagi LLM Benchmark75Accuracy (%)91.5
UGI - Writing55.03Writing Score90.8

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