GLM-5
used in
0
stacks
Zhipu AI's MIT-licensed 744B-parameter MoE model targeting complex systems engineering and agentic tasks.
GLM-5 is an open-weight mixture-of-experts model with 744B total parameters and 40B active per token, released under the MIT license in February 2026. The model card emphasizes tool usage and reasoning for agentic tasks, with evaluations on agent benchmarks such as Terminal-Bench 2.0 and BrowseComp. It uses DeepSeek Sparse Attention to reduce deployment cost while preserving long-context capacity, and is available through the Z.ai API platform and third-party inference providers.
Strengths
- MIT-licensed open weights
- Designed specifically for agentic and tool-use workloads
- Sparse attention architecture reduces long-context serving cost
- 40B active parameters keep inference cost moderate for its class
Tradeoffs
- 744B total parameters still demand multi-GPU infrastructure to self-host
- Smaller first-party API ecosystem than DeepSeek or Alibaba