Z.ai GLM-5.2 Matches US Models at Cybersecurity
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Z.ai's GLM-5.2 matches the latest US models at finding security bugs, per researchers cited in the Wall Street Journal.
- US export curbs on top AI like Anthropic's Mythos are drawing criticism, with WSJ's framing arguing Washington is handing Beijing a cyberwarfare advantage.
- Tokyo-based Sakana AI and China's 360 each launched cybersecurity models — Fugu and Tulongfeng respectively — claiming parity with Anthropic's banned Mythos and Fable 5.
- Chinese AI models are pressing into enterprise markets on price, with a Jefferies report (cited by China Money Network and Business Today) flagging cheaper Chinese alternatives challenging Western leaders.
- Crowd-sourced benchmarks are split: one tester reports GLM-5.2 (Max) ranks above Claude Opus 4.8 (Thinking) on Code Arena frontend tasks, while another X user publicly offered $100-to-$1 that GLM-5.2 will score below Mythos at UK AISI's cyber range.
- Internal benchmarking shared on X showed GLM-5.2 slightly outperforming Opus 4.8 on an internal mortgage servicing knowledge-base task at less than half the cost.
Why it matters: Z.ai's GLM-5.2 reportedly matching restricted Anthropic models on cybersecurity bugs undercuts the rationale for US export curbs, while Sakana AI and China's 360 launching rival models and a Jefferies report highlighting cheaper Chinese enterprise options compound the pressure — defenders argue the embargo now hands Beijing offensive capability at lower cost.
