glm-5 is an open-source Mixture-of-Experts large language model built for advanced coding and long-horizon agentic engineering tasks. It is designed to sustain optimization over hundreds of rounds and thousands of tool calls where previous models tend to plateau.
Project overview
It addresses the gap between model competence and real-world excellence in extended agentic workflows by maintaining effectiveness across long interaction horizons and large tool-call volumes.
Project type
AI Agent · AI Coding · Model Development · Model Runtime
Use cases
Coding & Development · Automation
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers and AI engineers who need an open-source model for extended agentic workflows involving advanced coding and many successive tool calls.
Key capabilities
Built to remain effective on agentic tasks over much longer horizons, sustaining optimization over hundreds of rounds and thousands of tool calls.
Limitations and risks
Because of the model size, deployment cost and hardware requirements may be high for standard users. GPU specifications are not documented, so prospective users should verify their hardware before adoption.
Getting started
Select a supported inference framework (for example vLLM, SGLang, or Transformers), download the model weights from Hugging Face or ModelScope, load the model, and interact with it through the framework's API or interface. Setup difficulty is moderate because it requires choosing and configuring an external inference framework and obtaining the weights.
Evidence and sources
GitHub project description: GLM-5: From Vibe Coding to Agentic Engineering
README: GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delive…
README: GLM-5.1 is our next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than its predecessor.