Coding GLM 5.1

Z.AI introduced GLM-5.1 as its flagship model for agentic engineering, positioning it as a purpose-built coding engine rather than a general-purpose conversational system. It is released as an open-weight model under the MIT license, which means organizations can self-host the weights, fine-tune for domain-specific codebases, and integrate the model into private development infrastructure without being tied to a single hosted endpoint. Independent commentary frames the family as a line of Chinese open-weight coding models optimized for complex systems design and long-running agentic workflows, with the 5.1 release representing a deliberate step up in autonomous, multi-step task execution.

At launch, GLM-5.1 took the top spot on SWE-Bench Pro and delivered a sizable jump over its predecessor on Terminal-Bench 2.0, two evaluations that stress repository-scale reasoning and end-to-end software engineering rather than isolated code completion. The model is described as capable of running autonomously for extended stretches, with one source citing up to eight hours of continuous agentic operation, making it a fit for refactors, repo exploration, and iterative engineering tasks that span many tool calls. Practically, developers can route Claude Code or other OpenAI-compatible coding interfaces through the BigModel endpoint, which lets teams adopt the model behind familiar tooling while still benefiting from its long-horizon coding behavior.