Nearly 500 tokens are sprayed every second! GLM-5.3 high-speed version quietly unveiled

Zhidongxi Author | Chen Junda Editor | Li Shuiqing Zhidongzhi reported on October 12 that recently, Zhipu quietly started testing the high-speed flagship model GLM-5.3 HighSpeed ​​in its AI programming tool ZCode. Leo, an AI programming blogger on the X platform, broke the news that he has grayscaled this model, and its output speed can exceed 490 tokens/s. What is the concept of 490 tokens per second? Currently, data from the large model API evaluation platform AI Ping shows that the output speed of the DeepSeek V4.1 Flash official API is 121.34 tokens/s, while the ZXQ official GLM-5.3 API speed is 70.84 tokens/s. In other words, GLM-5.3 HighSpeed ​​has reached a speed several times that of the current domestic mainstream models. GLM-5.3 HighSpeed ​​is currently only randomly open to ZCode subscribers, and there is no clear application entrance for internal testing. The rapid speed it showed in internal testing should only be used as a reference. In the end, it still depends on the speed when it is actually launched and provides services on a large scale to the outside world. ▲ZCode grayscale test GLM-5.3 HighSpeed ​​(Source: X Platform @leo114119) In fact, GLM-5.3 HighSpeed ​​appeared on the official website of the open source programming agent Pi harness a month ago. The input and output price of this model has not yet been disclosed. Judging from the basic information of the model, GLM-5.3 HighSpeed ​​has a context window of 1 million tokens and can output up to 131072 tokens at a time, which is consistent with the information of GLM-5.3. This is not the first time that Zhipu has launched a high-speed flagship model. In May of this year, GLM-5.1 launched a high-speed version of the API "GLM-5.1-highspeed" for some enterprise users, with a model output speed of 400 tokens/s. when