Chinese startup Z.ai, formerly known as Zhipu, has announced the release of GLM-4.5, an advanced open-source AI model family designed to excel in reasoning, coding, and autonomous agentic tasks. With a massive 355 billion parameters, GLM-4.5 integrates hybrid thinking to balance speed and task complexity, setting a new benchmark for AI performance in open-source offerings. According to Z.ai’s blog, GLM-4.5 is now ranked among the top open-source models globally, trailing only industry leaders like o3 and Grok 4. The model shines particularly in agentic tasks, outperforming rivals such as Gemini 2.5 Pro and Grok 4 in key benchmarks, with an impressive 90% success rate in tool use evaluations. What sets this release apart is not just the model’s performance but its accessibility. Z.ai has launched both GLM-4.5 and the lighter GLM-4.5-Air with open weights. Furthermore, the company has open-sourced its ‘slime’ training framework, enabling the AI community to replicate, experiment, and innovate on top of this cutting-edge foundation. This move is part of a broader trend among Chinese AI labs—including Qwen, DeepSeek, MiniMax, and Kimi—who are accelerating the development of frontier models and applying increasing pressure on established players like OpenAI. By continuously releasing competitive and affordable open models, Z.ai and its peers are reshaping the global AI landscape and making advanced AI capabilities more widely accessible.