
China’s $0.11 AI Bombshell: GLM-4.5 Shakes Up the Game for Pennies
A new wave is coming from the East—and it’s dirt cheap. Ambitious Chinese startup Z.ai has unveiled GLM-4.5, an open-source, ultra-efficient AI model that’s poised to upend the global AI balance of power. Compact, agile, and laughably inexpensive, it’s not just a technical marvel—it’s a strategic grenade lobbed into the heart of the AI race.
Running on just eight Nvidia H20 chips—not a futuristic supercluster, but a setup even indie developers could dream of—GLM-4.5 takes an agent-based approach, breaking down complex tasks into smaller, more accurate pieces. And if that sounds like a luxury service, the real kicker is the price.
Z.ai is charging just $0.11 per million input tokens and $0.28 per million output tokens. For comparison: DeepSeek R1 charges $0.14 and $2.19, while Moonshot’s Kimi K2, backed in part by Alibaba, asks $0.15 and a steep $2.50. GLM-4.5 isn’t just a genie in a bottle—it’s one that only charges for the cork.
Z.ai CEO Zhang Peng insists there are no supply issues with chips, and the compute is ready to scale. Training costs remain undisclosed, but the company promises transparency soon. The fact that it’s already drawing IPO speculation in China only adds intrigue.
And Z.ai didn’t come from nowhere: since 2019, it’s raised over $1.5 billion from giants like Alibaba, Tencent, and even Saudi Aramco–linked funds and Chinese municipal investors. The recent addition of Z.ai to the U.S. sanctions list feels less like punishment and more like a government-issued badge of relevance.
GLM-4.5 won’t be alone. China’s AI sector is on fire. Tencent has launched the 3D-generating HunyuanWorld-1.0, and Alibaba just dropped Qwen3-Coder for developers. China no longer looks like it’s trying to catch up—it looks ready to take the lead.