BREAKING
GLM-5.2 Rivals Claude Mythos on Bugs
IDOR Detection F1 Score
GLM-5.239
Opus 4.637
Opus 4.8/4.728
0$
per true positive
0B
total params
0M
token context
GLM-5.2 vs Claude Mythos
GLM-5.2Zhipu AI
Open weights, MIT license
Released June 16, 2026
Self-hosting and fine-tuning
Claude MythosAnthropic
Unreleased frontier model
Limited Project Glasswing access
Found thousands of zero-days
Coding Benchmark Scores
Terminal-Bench81
Opus 4.885
SWE-bench Pro62.1
One Task, Big Questions
AI NEWS BLITZ
An open-weight Chinese model is matching a top US frontier model at finding security bugs.