[Is Progress Stuck Without Claude? Overseas AI Blogger Proposes "Complete Stagnation" of Large-Scale Chinese Models] Beating AI News reports that overseas AI blogger Da7em has raised a controversial point: since Anthropic began strictly preventing other companies from distilling Claude, large-scale Chinese models seem to have collectively fallen silent. Previously, leading Chinese labs could release significantly stronger models every one or two months, but now, several months have passed without a comparable breakthrough. He cited the performance of four companies: - Kimi: The K3.1 Pro was released approximately 100 days after the K3 launch, but he believes the new model did not achieve a major breakthrough. - DeepSeek: Has yet to release a Pro-level flagship model that satisfies him. - Zhipu: Previously, they would release a new flagship model in less than two months, but now it has been about three months. - Alibaba: Most of the feedback he has heard about the Qwen4-Max preview version is disappointing. He attributes the performance of these companies to Anthropic's strengthened anti-distillation measures. In February of this year, Anthropic accused DeepSeek, Dark Side of the Moon, and MiniMax of using approximately 24,000 unauthorized accounts to make over 16 million calls to Claude to extract its reasoning, programming, and agent capabilities. In October, Anthropic disclosed that new safeguards had reduced the amount of reasoning content that unauthorized individuals could extract by approximately 90% to 99%. According to Da7em's inference, Chinese labs previously rapidly caught up to cutting-edge levels by distilling Claude and using a large amount of Claude's responses to train their own models. Now that this path is restricted, the model iteration speed has slowed down. Kimi and DeepSeek have turned to distilling OpenAI's GPT-6 Astra, but have not yet achieved the expected results. In his view, this further illustrates that Claude was the key source of capabilities upon which the rapid progress of Chinese models previously relied. In August of this year, Da7em also publicly refuted accusations that exaggerated the distillation of American models by Chinese models. In October, he even called the Kimi K3.1 Pro the best open-source model in the world. Now, he says his perspective has completely changed. He believes that while Chinese labs have excellent researchers, they are overly reliant on the modeling capabilities of other companies. Chinese models need to develop their own technological paths, rather than always serving as cheaper alternatives. [Original Link]