x.comthreadsteve hsu (@hsu_steve) on X
A tweet reporting that GLM-5.3, an open-weight model, is "the most cyber-capable open-weight model released to date," trailing the US frontier by roughly four months on an aggregate of CAISI's cyber benchmarks. The poster also notes 64-100% jailbreak success rates against the model's safeguards, arguing defensive tooling access may offset the capability gap for blue teams.
TL;DR GLM-5.3 tops open-weight models on CAISI's cyber benchmarks but is ~4 months behind the US frontier, with 64-100% jailbreak success rates undermining its safeguards.
Related in evals
