zai-org / GLM-5
View on GitHub →
Open weights · agentic foundation model

From vibe coding
to agentic engineering.

GLM-5 is built for the part of the loop that matters — planning, tool use, and following a task through to a working result.

01 · INTENT
Vibe
A loose prompt. A direction, not a spec — where most coding assistants stop.
02 · PLAN
Decompose
The model breaks intent into ordered, tool-using steps it can verify as it goes.
03 · OUTCOME
Engineered
A checked, working change — not a draft to clean up by hand.
open weights tool-use native long-horizon tasks

Stage labels illustrate the model's intended workflow; no benchmark figures shown.