Overview
- Zhipu AI published GLM‑5.2 as an open‑weight model on June 13, releasing core weights and formats that let developers download and run the system locally.
- Developer platforms and users moved quickly to adopt the model, with Vercel’s CEO praising its coding skill and the company integrating GLM‑5.2 into its AI Gateway within days of release.
- Public benchmark results show large single‑generation gains for coding tasks, with GLM‑5.2 placing near the top on Code Arena, Terminal‑Bench and FrontierSWE for front‑end and long‑horizon engineering tests.
- Technical caveats temper the enthusiasm: GLM‑5.2 uses a Mixture‑of‑Experts design with disputed parameter reporting, requires heavy quantization and hundreds of gigabytes of memory to run locally, and shows lower token efficiency on some tests and weaker results on certain private benchmarks.
- Investor interest and strategic concern followed the rollout, driving Zhipu’s Hong Kong market value above HK$1 trillion and prompting debate about lower‑cost open models, export controls, supply chains and how firms will choose between open and closed frontier systems.