CLAIM-2026-08-18-04

CyberGym rose from 77.2% to 84.5%.

Why it matters

Corroborates the direction of travel on a second, independent cyber benchmark, so the ExploitBench move is not a single-benchmark artefact.

Sources

  • z.ai/blog/glm-5.3
  • MarkTechPost
  • Unite.AI
  • Axios