In a significant leap forward for artificial intelligence, the rising Chinese AI company Z.ai, formerly known as Zhipu, has unveiled its groundbreaking gen-next base model, GLM-4.5. This release marks a new benchmark, achieving state-of-the-art (SOTA) performance in the realm of open-source models, underscoring Z.ai’s relentless commitment to pushing the boundaries of AI innovation and accessibility.

The GLM-4.5 model is distinguished by its fully self-developed architecture, positioning it as China’s most advanced open-source Mixture of Experts (MoE) model. This sophisticated design not only establishes new benchmarks in the AI world but also champions enhanced AI performance and accessibility while simultaneously reducing operational costs for developers globally, a pivotal step towards democratizing powerful AI capabilities.

Z.ai’s newest flagship model, GLM-4.5, is launched in two potent versions to cater to diverse needs: GLM-4.5, boasting an impressive 355 billion total parameters, and GLM-4.5-Air, with a robust 106 billion total parameters. Both iterations deliver integrated reasoning and advanced agentic capabilities within a unified framework, alongside specialized coding functionalities meticulously crafted for complex agent-based applications, expanding the scope of what developers can achieve.

Evaluated across twelve rigorous benchmarks, GLM-4.5 has garnered significant acclaim, ranking third globally among all AI models. Notably, it commands the top spot among all domestic and open-source models in terms of average performance score, solidifying its position as a frontrunner in the competitive landscape of artificial intelligence development.

A key advantage of the GLM-4.5 series lies in its superior speed and cost-efficiency. API calls for the model are remarkably affordable, starting at just $0.11 per million input tokens and $0.28 per million output tokens. Furthermore, its high-speed version supports impressive generation rates exceeding 100 tokens per second, making Z.ai’s new model one of the most economical and efficient solutions available to developers today.

Distinguishing itself further, GLM-4.5 is Z.ai’s inaugural foundation model to incorporate a native agentic design. This innovative architecture empowers the model to autonomously plan multi-step tasks, manage intricate full workflows, and generate complex visualizations. This intrinsic capability dramatically enhances the model’s versatility and utility across a myriad of applications, from intricate data processing to creative content generation.

Released under an open, auditable license, GLM-4.5 champions unprecedented control and transparency for enterprises, allowing for fine-tuning and seamless on-premise deployment. This open approach provides much-needed flexibility in an industry often dominated by proprietary systems. As Z.ai CEO Zhang Peng aptly highlights, “GLM-4.5 proves that cutting-edge AI can also be open, efficient, and affordable,” echoing the company’s vision for widespread AI adoption.

Z.ai, founded in 2019, has rapidly ascended to the forefront of AI development, with over 40 million downloads for its industry-leading models including the expansive GLM Series (6B-355B parameters), AutoGLM (China’s earliest agent model), and high-performing Flash Models. Recognized in Stanford’s 2025 AI Index Report and acknowledged by OpenAI for building competitive models, Z.ai’s global reputation continues to grow with the impactful launch of GLM-4.5, further solidifying its role as a pioneer in advanced artificial intelligence.