Overview
- Zhipu published GLM-5.2 as open weights under an MIT license and released the model on June 13, making it available on platforms such as Hugging Face and in GGUF format for local hosting.
- Developers and platforms moved quickly to adopt the model with Vercel integrating GLM-5.2 into its AI Gateway within days and several users calling it a viable daily driver for coding.
- Third-party and company benchmarks show large single-generation gains over GLM-5.1 and top placements on coding leaderboards, including a No. 2 rank on Code Arena for front-end web tasks.
- Technical caveats remain: Zhipu says GLM-5.2 uses a Mixture-of-Experts design with disputed total and active parameter counts, critics say private tests show gaps on some evaluations, and the model can be less token-efficient than Western rivals.
- The release has moved markets and policy debates by pushing Zhipu’s Hong Kong market value above HK$1 trillion and renewing questions about self-hosting economics, US export controls, and how enterprises choose between open models and paid APIs.