[Why Has DeepSeek Suddenly Become So Powerful? Foreign Media Reveals the Inside Story of Astra Distillation, Liang Wenfeng Compromises to Competitive Pressure] Beating AI News Flash: DeepSeek's new model has seen a significant increase in capabilities, and Astra's contribution may indeed be behind it. The Information, citing multiple sources with direct knowledge of the matter, reports that DeepSeek's V4.1-Flash-1017 update on October 17th, as well as the newly released Kimi K3.1 Pro from Dark Side of the Moon, both heavily utilized OpenAI GPT-6 Astra's output for distillation during training. DeepSeek's new version performs particularly well in browser operation tasks. In the Browser Use test with 180 real-world web page tasks, its score improved from 70.9 in the old version to 75.0, approaching Astra's 77.9. However, the model call cost per task is only $0.22, while Astra's is as high as $10.77, a difference of nearly 49 times. Previously, Claude was a primary target for distillation by domestic AI companies. However, Anthropic has tightened its defenses in the past two months, significantly increasing the difficulty of extracting the model's inference process. As a result, domestic labs have turned their attention to Astra. DeepSeek founder Liang Wenfeng was initially reluctant to rely on distillation in his V4 development. However, after seeing the progress made by domestic competitors like Zhipu GLM using distillation, he ultimately succumbed to competitive pressure. [Original Link]