Z.ai26 Aug 2026Model

GLM-5.3-Flash

Benchmark results

Benchmark	Score	Source check
Humanity's Last ExamWith tools · full set; 300,000-token context with context management; GPT-5.6 Luna judge	55.3%	Corroborated The same number is printed in 3 other lab documents: 1 2 3
AutomationBenchAutomationBench v1.0.6 with a fix for null-type handling	48.8%	Corroborated The same number is printed in 3 other lab documents: 1 2 3
DeepSWE v1.1mini-swe-agent harness; 6h timeout; 400K context	63.4%	Corroborated The same number is printed in 3 other lab documents: 1 2 3

Sources

Reported by Z.ai