ZDAI Launches GLM-5.3 Model to Enhance Cybersecurity

Chinese AI startup ZDAI announced that its open-source model, GLM-5.3, achieved an 84.5% success rate in the Cyber Game test, surpassing Anthropic's Mythos 5. GLM-5.3 differs from Mythos 5 in its ability to convert discovered vulnerabilities into effective attacks, scoring 54.4% in the Exploit Bench test compared to 78% for Mythos 5. Additionally, GLM-5.3 completed 105 attack development tasks in two hours, while Mythos 5 completed 181 tasks in the same timeframe.