An open model nearly matches the frontier at vulnerability discovery

Claim

Z.ai's GLM-5.3 nearly matched Mythos 5 on vulnerability discovery at 84.5% to 83.8%, though it trails badly at building exploits, and its cyber gains grew so fast that open weights will lag launch by two weeks, while DeepSeek shipped an MIT-licensed rival to Claude Code.

Reported metric: 84.5% vs 83.8%

Reporting date: 2026-08-15 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to August 15, 2026 (theinnermostloop.substack.com)