China's Knowledge Atlas Technology (2513), or Zhipu AI, will pivot its core focus toward Artificial General Intelligence —AI systems capable of matching or surpassing human intellect—rather than pursuing short-term commercialization, founder Tang Jie disclosed in an internal memo on Saturday.
Tang dubbed the strategic shift the "Touch High Plan." As the broader tech industry aggressively accelerates commercialization, Zhipu has decided to push upward to challenge the limits of physics and algorithms. "The giant wave has arrived, and the trend is irreversible," Tang wrote in the memo. "Anything short of the summit is failure."
According to Tang’s memo, Zhipu’s future research will be driven by four core engines:
First, long horizon tasks: enabling AI to transition from instant question-answering to managing large-scale, complex engineering projects.
Second, autonomous agent system: transforming AI assistants into "digital employees" capable of autonomous debate, collaboration, code review, and resource allocation to achieve autonomous-driving level productivity.
Third, fully self training: standardizing the creation of new knowledge through AI-to-AI self-play.
Fourth, ultimate security governance: implementing robust security constraints that scale alongside the model's growing capabilities.
Tang pointed out that the evolution of the "upper bound of intelligence" has a clear path: AI is completing the leap from perceptual intelligence to cognitive intelligence; machines are no longer just seeing and hearing, but are beginning to understand and reason, and the next step will be AGI .
Zhipu recently released its GLM-5.2 model, which is widely recognized as reaching the forefront of international model capabilities. The open-source model supports millions of truly usable contexts and continues to lead the industry in executing long-range tasks.
