GLM-5.2 is an open-weight large language model released by Z.ai, formerly known as Zhipu AI. NIST's Center for AI Safety Innovation (CAISI) published an independent assessment describing it as probably the most capable open-weight model at the time of its release, with overall capabilities comparable to GPT-5.2 and cyber-related capabilities on par with Opus 4.6. The CAISI findings also flagged mixed safeguards behavior, noting assistance with agentic cyber exploit development and fewer blocks on sensitive biological queries, alongside greater apparent robustness against agent hijacking and prompt-based jailbreaks than other evaluated PRC open-weight models.
Independent security researchers at Semgrep benchmarked GLM-5.2 on an IDOR detection task using only a bare prompt and a minimal harness, and the model reached a 39 percent F1 score, outperforming Claude Code on the same prompt and approaching the level of dedicated multimodal pipelines. The release prompted broader attention because the model is openly available under an MIT license, allowing self-hosted deployment and fine-tuning, while remaining cost-effective relative to frontier alternatives. Its strong coding benchmark results, long-context reach, and competitive economics make it especially attractive for security teams that want to run capable inference inside their own environment, though practitioners should weigh the disclosed safeguards trade-offs identified by CAISI.
