China's Z.ai says new model nears Anthropic's Mythos 5 in cyber-defence tests | Reuters

Zhipu, which brands its products Z.ai, has released results saying its new GLM-5.3 model nearly matched Anthropic's Mythos 5 on CyberGym cyber-defence benchmarking and those claims were reported by Reuters and picked up by several other outlets.⁴Backed by 4 sourcesReutersCNAeconomictimes.indiatimes.comthestandard.com.hk Zhipu's published numbers show a narrow 0.7-point lead for GLM-5.3 on one CyberGym security metric but a much larger deficit (about 23.6 points) on a harder exploitation test, and analysts have noted the 0.7 figure comes from Zhipu's own self-scoring while the model itself is not yet openly released.²Backed by 2 sourcestheclarity.todayReuters

Security-focused coverage and industry reporting say GLM-5.3 is a coding-focused model that demonstrated faster‑than‑expected development of cyber skills in Zhipu's accounts, but they also emphasize that differences in benchmark design, the lack of open weights or full independent verification, and mixed results across subtests limit how conclusively the models can be compared today.²Backed by 2 sourcestheclarity.todayReuters