GLM-5 is Z.ai's (formerly Zhipu AI) flagship open-source foundation model, built on a 744-billion-parameter Mixture-of-Experts architecture with 40 billion active parameters. It is engineered for complex systems design and long-horizon agent workflows, delivering production-grade performance on large-scale programming tasks that rivals leading closed-source models. With advanced agentic planning, deep backend reasoning, and iterative self-correction, GLM-5 moves beyond code generation to full-system construction and autonomous execution.
API|Reasoning|Open ModelMIT
Knowledge Cutoff
2026-02-17
Input → Output Format
Context Memory
80KIN131KOUT
AI Performance Evaluation
Arena Overall Score
1457
±5As of 2026-04-23
Overall Rank
No.28
17,733 Votes
Arena by Ability
Hard Prompts
1478±6No.29
Expert Knowledge
1482±16No.28
Instruction Following
1446±8No.30
Conversation Memory
1467±11No.30
Creative
1447±11No.17
Coding
1493±9No.40
Math
1447±17No.33
Arena by Occupation
Creative Writing
1444±9No.26
Social Sciences
1482±11No.22
Media
1433±10No.24
Business
1451±10No.32
Healthcare
1493±17No.17
Legal
1467±16No.29
Software
1486±8No.37
Mathematics
1461±19No.24
Source:Arena Intelligence
Overall
AA Intelligence Index
50%↑11%
LiveBench
69%↑8%
ForecastBench
60%↑1%
Reasoning & Math
GPQA Diamond
82%↑1%
HLE
27%↑10%
LB Reasoning
69%↑9%
LB Math
83%↑10%
LB Data
68%↑18%
Coding
AA Coding Index
44%↑10%
LB Coding
74%↑0%
LB Agentic
55%↑12%
TAU2
98%↑25%
TerminalBench
43%↑12%
SciCode
46%↑5%
Language & Instructions
IFBench
72%↑16%
AA-LCR
63%↑2%
Hallucination (HHEM)
10%↑0%
Factual (HHEM)
90%↑0%
LB Language
78%↑6%
LB IF
55%↑9%
Output Speed
Standard Mode
52tok/s↓30
First Output 1.34s
Reasoning Mode
59tok/s↓29
First Output 60.38s
