Chinese AI startup Zhipu AI has launched its new-generation open-source large language model series, GLM-4.5 and GLM-4.5 Air, designed to unify advanced reasoning, coding, and agentic capabilities. The flagship GLM-4.5 model features 355 billion total parameters with 32 billion active parameters using a mixture-of-experts (MoE) architecture, while the lighter GLM-4.5 Air has 106 billion total parameters with 12 billion active. These models support two operational modes: a "thinking mode" for complex reasoning and tool use, and a "non-thinking mode" for instant responses. GLM-4.5 has demonstrated competitive performance against leading global AI models such as Claude 4 Opus and Gemini 2.5 Pro, excelling particularly in coding, reasoning, and agentic tasks. The models are offered under an MIT license with API pricing at $0.6 per million input tokens and $2.2 per million output tokens for GLM-4.5, and lower rates for GLM-4.5 Air. The launch marks a milestone in China's AI development, with GLM-4.5 topping domestic evaluations and matching global top models while costing approximately one-tenth of Claude's API fees. The models have been trained on a 22 trillion-token corpus and incorporate advanced features such as GQA, partial RoPE, multi-token prediction, Muon optimizer, and native tool use capabilities. Meanwhile, other Chinese open-source models such as Qwen3 series continue to advance, with Qwen3 235B 2507 and Qwen3-Coder-30B-A3B models gaining traction for reasoning, coding, and agentic benchmarks, some offering local deployment with large context windows up to 1 million tokens. Additionally, Cerebras Systems has introduced hosted endpoints for Qwen3 models, providing fast inference speeds up to 2,000 tokens per second with subscription plans starting at $50 per month. This surge in Chinese open-source AI models reflects an ongoing acceleration in frontier AI development with multiple models achieving top rankings on international leaderboards and competitive performance against established Western AI systems.