According to monitoring by Dongcha Beating, the Zhiyu AI open source model GLM-5.2 has officially entered the long-range software engineering benchmark DeepSWE. In maximum reasoning mode, the success rate for complex development tasks reached 44%, ranking first among open source models. This is a 13 percentage point increase compared to the previously listed Kimi K2.7 Code. The average cost to solve each task with GLM-5.2 is $3.92, slightly higher than Kimi K2.7 Code's $2.82, yet its success rate surpasses that of several mainstream closed-source models under specific reasoning configurations, including Claude Sonnet 4.6 [high] (30%), Gemini 3.5 Flash [medium] (37%), and Claude Opus 4.8 [low] (41%). The DeepSWE benchmark, designed by Datacurve, specifically tests the ability of AI agents to solve long tasks. The tests include 113 real programming problems across 5 languages. Unlike traditional tests that only modify a single piece of code, DeepSWE requires AI to collaboratively modify multiple files, averaging over 600 lines of code for repairs. The evaluation runs in isolated containers, strictly limiting CPU and memory resources.