Z.AI, the Beijing-based company formerly known as Zhipu AI, has publicly released its latest large language model, GLM-5.2. The model weights are immediately available on Hugging Face under a permissive MIT license.
GLM-5.2 features a 1 million-token context window, a fivefold increase over its predecessor. This capacity places it among a rare group of open-weight models capable of processing book-length documentation or entire codebases in a single pass.
Built on a Mixture of Experts (MoE) architecture, the model holds up to 753 billion total parameters but activates only about 40 billion at a given time. The release includes an FP8 reduced-precision variant to lower computational requirements.
Z.AI is positioning GLM-5.2 strictly as a coding and engineering tool, optimized for project-level workflows and long-horizon agentic tasks. It marks the third major release in the GLM-5 series over the last four months.
The model is accessible to GLM Coding Plan subscribers, with tiers starting at $18 per month. Direct API access is priced at $1.40 per million input tokens and $4.40 per million output tokens. No third-party performance benchmarks were published alongside the launch.