Z.ai has recently and officially open sourced GLM-4.5! This new cutting-edge advancement in open-source base models is a hallmark development in today’s artificial intelligence landscape. This new model sets a performance benchmark for open foundation models and ranks third globally, while proudly holding the top position among open-source and Chinese-developed models.
GLM-4.5 is very important because it is China’s first open-source Mixture of Experts (MoE) model. Most importantly, it is proudly built on a completely internally developed stack. This new, creative framework aims to provide superior performance while significantly lowering deployment costs for developers.
GLM-4.5 is available in two configurations: the flagship model featuring 355 billion parameters and a lighter version called GLM-4.5-Air, which operates with 106 billion parameters. This freedom of choice lets developers select the model that meets their needs without sacrificing performance for their use case.
One of the most impressive features in GLM-4.5 is its fast inference time, exceeding 100 tokens per second. This efficiency is reflected in our highly competitive pricing. We only bill $0.11 / million input tokens and $0.28 / million output tokens. This aggressive pricing strategy makes GLM-4.5 a compelling value compared to other purpose-built AI accelerators across the full array of applications.
The new features that the model has such as native reasoning, coding, and agentic functionalities. This featured capability builds its flexibility for developers creating across all recommended industries. Z.ai also thoroughly tested GLM-4.5 across 12 benchmark tests to prove its consistent performance and dependable performance.
Since launching in 2020, Z.ai has quickly taken off. Its open-source models, like GLM-4.5, recently broke the 40 million download milestone, which is a clear sign of the growing interest and trust in their technology. Z.ai has received accolades from OpenAI for being one of the handful of global companies that are capable of building competitive AI models. Moreover, it has won over 50 industry awards for its innovative implementations.
