GLM-5.3 Nears Mythos 5 in Cybersecurity

Chinese company Z.ai released a high-speed version of its open-source model GLM-5.3, designed for use in cybersecurity. The model achieved advanced performance in the CyberGym test, scoring 84.5%, surpassing the Mythos 5 model, which scored 83.8%. However, its performance was lower in the ExploitBench test, with a score of 54.4% compared to 78% for Mythos 5. The model demonstrated the ability to complete 105 tasks in two hours and 130 tasks in six hours, versus 181 and 247 tasks for Mythos. Initially, the model's capabilities will be limited to trusted staff with access, along with strengthened security measures, aiming to support defensive applications and cybersecurity efforts—an attempt to address the challenges posed by sharing open-source AI models.

articleDetails.fontSize.sizeحجم النص

تنويه: هذا ملخص تم إنشاؤه بواسطة الذكاء الاصطناعي

عرض المقال بالكامل