GPT-6 Astra Distillation: DeepSeek, Moonshot Named in Report

A new report suggests that the Kimi K3.1 Pro and GLM-5.3 models were developed through a process of distillation from OpenAI's GPT-6 Astra. This practice involves training smaller, more efficient AI systems by utilizing the outputs of larger, more powerful foundation models. The findings highlight ongoing industry discussions regarding the origins of various competitive language models and the technical methodologies used to achieve performance gains.