GLM-5.3 Nears Claude Mythos Exploits

The GLM-5.3 model has made significant strides in vulnerability exploitation tests, nearly matching the performance of Claude Mythos.

According to Andrew's Letter in The Batch, the GLM-5.3 open weights model scored 12% against Claude Mythos' 14% in these tests.

The rapid gains in open weights model performance across the AI industry are underscored by Xiaomi's new open weights leader and other recent developments.