Pushing boundaries, Z.ai sets a new benchmark for open-source AI performance and affordability.
In a new development in the world of artificial intelligence (AI), sources reveal how a rising Chinese AI company, Z.ai (formerly Zhipu), has released a gen-next base model GLM-4.5, achieving SOTA performance in open-source models. The new development exemplifies the company’s commitment to pushing boundaries, setting a new benchmark for open-source AI performance and affordability.
Z.ai’s newest flagship model, GLM-4.5, is said to have been built on a fully self-developed architecture, rising as China’s most advanced open-source Mixture of Experts (MoE) model. This has paved the way for newer benchmarks in the AI world, thriving on AI performance and accessibility while also reducing costs for developers.
GLM-4.5 emerges as one of the most talked-about innovations by Z.ai as its first open-source MoE model and is available in two powerful versions, namely GLM-4.5 with 355 billion total parameters and GLM-4.5-Air with 106 billion total parameters. Both models deliver integrated reasoning and agentic capabilities within a unified framework, along with coding designed specifically for complex agent-based applications. It has already been evaluated across twelve rigorous benchmarks, and GLM-4.5 ranks third globally, taking the top spot among all domestic and open-source models in terms of average score.
A few of the essential advantages of the GLM-4.5 series include higher speed and cost advantages, where API calls start at just $0.11 per million input tokens and $0.28 per million output tokens. On the other hand, the high-speed version supports generation rates over 100 tokens per second. These features make the new model by Z.ai one of the most affordable and efficient models that are available to developers today.
Z.ai with GLM-4.5 also stands different because it is the company’s first foundation model to feature native agentic design. This architecture allows the model to autonomously A. plan multi-step tasks, B. manage full workflows, and C. generate complex visualizations. The new model has been released under an open, auditable license, which allows for fine-tuning and on-premise deployment. This offers enterprises with the much-needed control and transparency in a field that is dominated by proprietary systems. CEO of Z.ai, Zhang Peng, on the new model highlights, “GLM-4.5 proves that the cutting-edge AI can also be open, efficient, and affordable.”
Z.ai has been recognized in Stanford’s 2025 AI Index Report and named by OpenAI as one of the few global firms building competitive models. These international acknowledgments reflect the company’s growing reputation ahead of the launch of GLM-4.5.
Z.ai was founded in 2019, and today it’s at the forefront of AI development. With more than 40 million downloads, its industry-leading models include GLM Series (6B-355B parameters), AutoGLM (China’s earliest agent model), Flash Models (lightweight, high-performing solutions), Model-as-a-Service (MaaS – scalable AI deployment), and Agent Ecosystem (build-your-own assistants).
