GLM 4.7

GLM 4.7 – Agentic Coding MoE with Three-Mode Thinking

GLM 4.7 is Z.AI's (Zhipu AI) open-source Mixture-of-Experts language model built around two explicit upgrade axes: enhanced programming capabilities and more stable multi-step reasoning and execution. It introduces a structured three-mode thinking system not present in prior GLM releases.

Three-Mode Thinking System

GLM 4.7 introduces three explicitly documented and independently controllable reasoning modes:

• Interleaved Thinking: The model thinks before every response and every tool call, improving instruction following and code generation quality.
• Preserved Thinking: In agent coding scenarios, reasoning blocks are automatically retained across multi-turn conversations — reusing prior reasoning rather than re-deriving from scratch, reducing information loss on long-horizon tasks.
• Turn-level Thinking: Reasoning can be toggled per-turn within a single session — disabled for lightweight requests to reduce latency and cost, enabled for complex tasks to boost accuracy.