BREAKING
Zhipu GLM-5.2 Matches Anthropic in Cyber
0
B
parameters
0
M
context tokens
IDOR Detection F1: 39% vs 32%
GLM-5.2
39
Claude Code
32
SWE-bench Pro: 62.1 vs 58.6
GLM-5.2
62.1
GPT-5.5
58.6
Low-Cost Power Raises Governance Gaps
Strengths
●
About $0.17 per vulnerability
●
Runs locally, MIT license
●
Agentic cyber workflows
Concerns
●
Hard to enforce logging
●
Reports of misuse attempts
●
Trails on general tasks
US-China Frontier Gap Narrows Fast
AI NEWS BLITZ
China's Zhipu has released GLM-5.2, an open-weight model rivaling Anthropic on cyber tasks.