Zhipu drops GLM-5, a Chinese model nipping at Claude Opus 4.5's heels
Another open-source model catching up to the top tier, but this one was trained on Huawei Ascend chips, not Nvidia. If you code with AI, nothing changes for you today. But it confirms that dependence on American chips is no longer a blocker for Chinese labs.
practitioners › GLM-5, open source, performance close to Claude Opus 4.5 on coding and beating Gemini 3 Pro on some benchmarks, trained on Huawei Ascend chips, released February 11, 2026.
Reuters →