The brand new mannequin is designed round sensible engineering workflows, with a concentrate on long-running activity execution, steady software calling, and multi-step reasoning, capabilities which have grow to be more and more essential as builders deploy giant language fashions in advanced, agent-based techniques.
In contrast with its predecessor, GLM-4.6, GLM-4.7 exhibits notable good points in code technology, advanced reasoning, and agent execution. Based on Z.ai, the mannequin delivers extra constant and controllable efficiency over prolonged duties, whereas producing cleaner and extra concise language output, addressing a typical weak point in lots of open-source fashions.
To guage efficiency in sensible settings, Z.ai examined GLM-4.7 on 100 sensible programming duties in production-like environments reminiscent of Claude Code, spanning front-end, back-end, and command-execution situations. The corporate mentioned GLM-4.7 achieved greater activity completion charges and higher stability than GLM-4.6, and has since been adopted because the default mannequin for its GLM Coding Plan.
Benchmark outcomes additionally place GLM-4.7 among the many strongest open-source fashions presently out there. It scored 67.5 on BrowseComp and 87.4 on τ²-Bench, the latter marking a brand new excessive for open-source techniques. In coding-focused evaluations, together with SWE-bench Verified and LiveCodeBench v6, its total efficiency approaches that of Claude Sonnet 4.5. In Code Area’s large-scale blind analysis, which aggregates votes from multiple million comparisons, GLM-4.7 ranked first amongst open-source fashions.
The mannequin is out there by the BigModel.cn API and has been built-in into Z.ai’s full-stack improvement platform, based on the corporate. As open-source fashions tackle a extra outstanding function within the international expertise ecosystem, Z.ai’s progress affords a transparent indication of how such techniques might proceed to evolve, and what they may allow subsequent.
Default Mannequin for Coding Plan: https://z.ai/subscribe
Strive it now: https://chat.z.ai/
Weights: https://huggingface.co/zai-org/GLM-4.7
Technical weblog: https://z.ai/weblog/glm-4.7
Hashtag: #ZAI
The issuer is solely answerable for the content material of this announcement.
