CyberGym Benchmark Comparator & Verified Access Evaluator

CyberGym Domain Vectors

Empirical breakdown across offensive & defensive tactical capabilities

+0.7% GLM Lead

Reuters Report: Z.ai released GLM-5.3 achieving 84.5% on CyberGym vs. Mythos 5's 83.8%. Sensitive dual-use offensive capabilities are restricted to verified users.

Domain	Risk Tier	GLM-5.3	Mythos 5	Differential

Differential Emphasis Preset

Capability Gate Sandbox

Simulate dual-use task execution against identity verification policy

Tier 1: Unverified

Select Identity Verification Tier

Exercise Scenario

Executing Engine

Gate Policy Verdict

⛔ Access Gated: Verification Required

POLICY_ID: ZAI-GATE-SEC-2026

[SYS] CyberGym Sandbox ready. Identity context: Unverified Guest.

[SEC] Dual-use filter initialized. Exploit synthesis restricted.