Without Claude, there can be no progress? Overseas AI bloggers raise the theory of "comprehensive stagnation" of Chinese large models.

Beating AI News Flash: Overseas AI blogger Da7em has raised a rather controversial viewpoint: ever since Anthropic began strictly guarding against other companies distilling Claude, Chinese large models seem to have collectively gone silent.

Top Chinese labs used to release noticeably stronger models every one or two months, but now several months have passed, and they have yet to deliver breakthroughs of a comparable magnitude.

He cited the performance of four vendors:

- Kimi: It took about 100 days after the release of K3 for K3.1 Pro to arrive, but he believes the new model has not achieved a major breakthrough.

- DeepSeek: It has yet to release a Pro-level flagship model that satisfies him.

- Zhipu: It used to launch a new flagship in less than two months, but now about three months have passed.

- Alibaba: Most of the feedback he has heard about the Qwen4-Max preview has been disappointing.

He attributes the performance of these vendors to Anthropic's strengthened anti-distillation measures.

In February this year, Anthropic accused DeepSeek, Moonshot AI, and MiniMax of using about 24,000 violating accounts to make more than 16 million calls to Claude in order to extract its reasoning, coding, and agent capabilities. In October, Anthropic also disclosed that new safeguards had reduced the reasoning content unauthorized parties could extract by about 90% to 99%.

According to Da7em's inference, Chinese labs previously used distillation of Claude to train their own models on large amounts of Claude responses, rapidly catching up to the frontier. Now that this path is restricted, the pace of model iteration has slowed accordingly.

Kimi and DeepSeek have turned to distilling OpenAI's GPT-6 Astra, but still have not achieved the expected results. In his view, this further shows that Claude was the key source of capability on which the rapid progress of Chinese models previously depended.

In August this year, Da7em also publicly opposed exaggerated accusations that Chinese models distill American models. In October, he even called Kimi K3.1 Pro the best open model in the world.

Now, he says what he has seen has completely changed his judgment. He believes that Chinese labs have excellent researchers but rely excessively on other companies' model capabilities. Chinese models need to develop their own technological path rather than always serving as cheaper alternatives.