On the evening of September 13, AI company Zhipu announced a major funding round of approximately $5 billion. The funds from this round will be primarily used for the development of the next-generation GLM foundation model, the construction of a fully self-training system, and related computing infrastructure.

In its specific R&D roadmap, Zhipu has clearly identified "fully self-training" as the core goal for the next-generation GLM model. This means that the next-generation GLM will be trained within the intelligent environment built by the previous generation of GLM, thereby forming a recursive self-improvement cycle. To achieve this technological advancement, the investment will comprehensively cover automated generation and screening of training data, the construction of task environments, and the enhancement of the model's long-range reasoning capabilities; meanwhile, engineering efforts will continue to promote compatibility with domestic chips, in-depth operator development, and optimization of inference efficiency.

Zhipu's current R&D strategy simultaneously addresses two key variables in foundational models: "model capability" and "computing efficiency." Through the fully self-training mechanism, the model deeply participates in building training resources, which can effectively expand the data and task space required for subsequent technological iterations. Meanwhile, chip compatibility and inference optimization focus on maximizing the effective output under the same hardware conditions.

The simultaneous advancement of these two technical paths marks that Zhipu is not only expanding the scale of model training but also actively exploring more efficient technological iteration paradigms. With the completion of this $5 billion funding round, Zhipu has successfully established a complete R&D foundation for the evolution of its next-generation model, spanning from the algorithmic system to the underlying engineering facilities.