GLM-5.2: Built for Long-Horizon Tasks
GLM-5.2 is a new AI model designed specifically for long-horizon tasks that require sustained reasoning and decision-making over extended periods. The model aims to improve performance on complex, multi-step problems by maintaining coherence and context across longer sequences of actions or thoughts.
Background
- GLM-5.2 is an open-weight language model from Zhipu AI (Beijing, China), the team behind the GLM/GLM-4 series and the ChatGLM product.
- "Long-horizon tasks" refers to multi-step reasoning, agentic workflows, and code generation that require maintaining coherence across very long contexts. The model reportedly supports a 128K-token context window out of the box and up to 2M tokens with tree-based positional encoding.
- Zhipu AI claims that on benchmarks like LiveCodeBench and AgentBench, GLM-5.2 outperforms comparable open models (e.g., Qwen3, DeepSeek, Llama 3) in coding, agent planning, and instruction following.
- The model is released under a permissive license for research and limited commercial use, continuing Zhipu's strategy of publishing competitive open-weight models against a backdrop of US export controls on advanced AI chips to China.