Zhipu AI's GLM-5.2 closes in on closed-source leaders in coding marathons

In three linesZhipu AI releases GLM-5.2 under MIT license with stable 1-million-token context. On FrontierSWE benchmark for long-duration coding tasks, the open-source model trails Anthropic's Claude Opus 4.8 by just one percentage point. Significant gap remains on reasoning versus closed-source rivals.

Summary generated by Claude — human-verified