China's Cheap AI Models Win Agentic Price War

Chinese AI labs Z.ai and Moonshot have launched new models that are nearly as intelligent as top competitors, but much cheaper. Silicon Valley startups are already using Z.ai's GLM-5.2. What's interesting is how high prices are putting Western models at a disadvantage. Agentic AI, which is more expensive to use than large language models, values the cheapest intelligence. Right now, China has the advantage. AI prices are calculated per token, a computing unit. Z.ai charges US$1.92 per million output tokens for GLM-5.2, while Anthropic charges $25 per million for its Opus 4.8. These cost differences are compounded by AI agents, which can perform online tasks and require dramatically more tokens. Recent estimates suggest agentic tasks use up to 3,500 times as many tokens as a simple reasoning task. The bottom line is that the emergence of agentic AI is driving a huge demand for tokens, and cheaper models are attracting institutions with high usage. This could significantly impact the future of AI development and accessibility.