Zhipu AI has open-sourced the GLM-4.5 series, a set of foundation models designed for intelligent agents that unify reasoning, coding, and tool usage capabilities. The release includes both the full-scale GLM-4.5 and the more compact GLM-4.5-Air, available under the MIT license for commercial use.

  • GLM-4.5 features 355 billion total parameters with 32 billion active parameters, while GLM-4.5-Air has 106 billion total and 12 billion active parameters.
  • Both models are hybrid reasoning architectures offering a "thinking mode" for complex tasks and a "non-thinking mode" for immediate responses.
  • The models achieve a score of 63.2 across 12 industry-standard benchmarks, ranking third among proprietary and open-source models, while GLM-4.5-Air scores 59.8.
  • Base models, hybrid reasoning versions, and FP8 quantized variants are available for download on Hugging Face and ModelScope.

The release provides developers with open weights to build complex agentic applications locally or via inference providers like vLLM and SGLang.