GLM-4.7

This GLM-4.7 configuration emphasizes the model's hybrid thinking behavior for reasoning-intensive tasks. Built on 358 billion total parameters (32 billion active) with a 200K context window and up to 131K output, it leans on the interleaved, preserved, and turn-level thinking mechanisms that produce reasoning before responses and tool calls and let developers control reasoning depth per turn. The model delivers frontier-level results in coding, terminal automation, and mathematical reasoning while exposing its intermediate reasoning chains. Its bilingual English and Chinese capabilities and MoE efficiency make it a fit for workflows where understanding the model's decision process matters as much as the final answer. Released under an MIT license, this thinking-forward configuration suits developers building interpretable reasoning pipelines, advanced debugging assistants, and complex multi-step planning, where transparent intermediate steps improve trust and control.