Zhipu's Next-Generation AI Model Expected to Feature Up to 3.5 Trillion Parameters

Zhipu, an artificial intelligence company, is reportedly developing a next-generation foundational model with a parameter count ranging from 2.5 trillion to 3.5 trillion. This information, derived from an expert interview circulating within the investment community, contradicts earlier estimates that suggested the model would have 1.6 trillion parameters. The current model, GLM-5.3, released in August 2026, utilizes a previous generation foundational model with approximately 744 billion parameters, indicating that the new model could be 3.4 to 4.7 times larger.

The training process for the new model has faced challenges due to Zhipu's adoption of a more aggressive model architecture. Reports indicate that there was a period during pre-training when the loss metrics did not improve, suggesting that significant computational resources were being utilized without progress. However, it is believed that these issues have been resolved recently, and training is expected to conclude soon, with the new model anticipated to reach performance levels comparable to Claude Fable.

In addition to its scale, the new model is designed with inference speed and cost as critical factors from the outset. This approach aims to ensure that the model can provide large-scale services immediately upon release. Furthermore, Zhipu is in discussions with multiple cloud vendors regarding revenue sharing arrangements to address potential supply challenges associated with the deployment of large AI models.

© 2026 KLEA News. All Rights Reserved. This article is provided for informational purposes only. It is not offered or intended to be used as legal, tax, investment, financial, or other advice.

Source: KLEA News