Zhipu AI (02513.HK) launched and open-sourced GLM-5.2 under an MIT license with no geographic restrictions. The firm says GLM-5.2 ranked first among globally available models on CodeArena, a front-end evaluation system with over 1 mln users in blind tests. The model features a 1M-token context window for long-range tasks, improved coding ability, and infrastructure optimizations for Day‑0 operation on domestic compute. Zhipu completed Day‑0 inference adaptation on multiple Chinese compute platforms, citing integration with Huawei Ascend, T-Head (Pingtouge), Moore Threads, Cambricon, Kunlunxin, Muxi, Hygon (Haiguang) and Biren, and reports high throughput, low latency and large-concurrency stability on domestic chip clusters. Benchmarks cited place GLM-5.2’s long-range performance between Claude Opus 4.7 and 4.8 and flag it as the highest-ranked open-source model. Zhipu expects Huawei’s Ascend 950 supernode, slated for H2, to become an additional compute base for GLM-5.2.(金十数据)