Recently, the market has been abuzz with major news that Zhipu plans to officially launch its new generation foundation large model, GLM-5.5, in August of this year. It is reported that this model will achieve a significant breakthrough in architecture, with the number of parameters expected to exceed 1 trillion.

The background of this upgrade stems from recent dramatic changes in the international large model landscape. Previously, U.S. authorities imposed access restrictions on two top-tier models from Anthropic—Fable5 and Mythos5—causing these models, which had long held the top positions on global rankings, to be temporarily unavailable for widespread use. In this context, Zhipu quickly released the GLM-5.2 version, based on an upgraded GLM-5.0 architecture, and made it fully open source for the developer community.

8月有望追赶美国最强神话大模型 智谱GLM-5. 5 将升级万亿参数

GLM-5.2 has rapidly climbed to high positions on various AI technology rankings due to its outstanding performance. Data shows that the model not only leads the open-source SOTA (State of the Art) list, but also demonstrates strong competitiveness in hard metrics such as AI programming capabilities, making it currently the first choice among the global developer community for "programmable large models."

The upcoming GLM-5.5 is seen as a strategic product for Zhipu to challenge the highest global performance levels. The current version of GLM-5.2 has approximately 744 billion parameters, with an activated parameter count of around 40 billion; if it indeed upgrades to a trillion-parameter scale, it would mean an increase of over 50% in the model's overall capacity. Industry experts generally believe that, combined with Zhipu's deep experience in large model post-training optimization, GLM-5.5 may become a crucial turning point for Chinese large models to catch up with global top levels.

Previously, in response to Elon Musk, CEO of Tesla, who stated that "Chinese large models still need more time to catch up with global top levels," Zhipu's founder Tang Jie gave an encouraging response, hinting that the pace of catching up would be faster than expected. As the release window in August approaches, the market is closely watching whether GLM-5.5 can achieve this goal as planned, thereby gaining new voice in domestic artificial intelligence technology.

From the rapid replacement of GLM-5.2 to the technical leap of GLM-5.5, Zhipu is accelerating the construction of the domestic computing power ecosystem through continuous iteration and open-source strategies. Although this process involves significant pressure from computing power investment and technological breakthroughs, the pace of large model localization has become unstoppable with the explosive growth of application-side calling demand.