On June 17, Zhipu AI announced the official launch and open source of its new generation flagship model GLM-5.2. This model is specially built for long-range mission capabilities. Its core features include 1M lossless context, enhanced coding capabilities, Day 0 adaptation to domestic computing platforms, and adoption of the MIT open source protocol.
According to the official introduction of Zhipu, GLM-5.2 achieved the first performance among available models in the world on Code Arena, a front-end development and evaluation system where millions of users around the world participated in blind tests. In multiple authoritative reviews such as FrontierSWE and Terminal-Bench, the gap between GLM-5.2 and the overseas head model Claude Opus 4.8 has narrowed to 1% to 4%, making it the highest-ranking open source model. Among them, GLM-5.2 was only 1% lower than Opus 4.8 in the FrontierSWE test and exceeded GPT-5.5; it scored 81.0 in the Terminal-Bench 2.1 evaluation, which was a significant increase of 17.5 percentage points from the 63.5 of the previous generation GLM-5.1.
GLM-5.2 implements a truly usable 1M context window, which can fully host large-scale software engineering projects in a long-term task. In the actual test, the model completed the complete application development and delivery covering the Web, mobile terminals and small programs at one time, processing a total of 880,000 tokens, and almost used up the 1M window. In the SWE-bench Pro evaluation, GLM-5.2 scored 62.1.
The model has been opened to all GLM Coding Plan users. Online inference will be adapted to domestic computing platforms such as Huawei Ascend, Pingtouge, Moore Thread, Cambrian, Kunlun Core, Muxi, Haiguang, and Biren on Day 0. GLM-5.2 weights have been launched on Hugging Face and ModelScope. They are open source under the loosest MIT license and can be freely downloaded, deployed and commercially used. The API is simultaneously launched on the BigModel open platform and Z.ai.