| Mobile Web

Kakao language model Kanana-2 outperforms overseas models in safety assessment

Kakao has disclosed numerical safety verification results for its self-developed lightweight language models. The company said on Aug. 18 that two open-source models in its Kanana-2 series scored higher than major global models of similar size on assessments of harmfulness and bias, confirming top-tier safety. The evaluation used AssurAI, a Korea-designed safety benchmark, and compared the models with Google’s Gemma and Alibaba’s Qwen.