The news: Z.ai’s new open-source GLM-4.5 model is undercutting DeepSeek and US rivals in cost and efficiency and intensifying global AI competition. 

• Available through GitHub or Hugging Face and also from chat.z.ai on browsers, the model has the potential to upend AI pricing models, per CNBC.
• GLM-4.5 is proficient in reasoning, coding, and multimodal and agentic tasks. Early demand surged, maxing out capacity by 10am EST Monday, per Z.ai’s X post.

Making more with less: GLM-4.5 costs just $0.11 per million input tokens and $0.28 per output—136 times cheaper on input and 268 times cheaper on output than Anthropic’s Claude 3 Opus, whose performance is similar.

The cost paradox widens: As US AI firms chase scale—training ever-larger models on sprawling, power-hungry data centers—Chinese startups continue to flip the script. 

While OpenAI, Anthropic, Meta, and Google push multimillion-dollar training runs and race to secure next-gen data centers, China’s AI players are compressing costs while ramping up performance. 

Models like Z.ai’s GLM-4.5, DeepSeek’s R1, and Alibaba’s Qwen3, which set open-source records last week, are already demonstrating efficiency over cost.

How long can US firms justify rising expenses when open-source rivals continuously deliver “good enough” performance at a fraction of the cost?

Our take: For marketers, open-source tools like Z.ai offer affordable alternatives to costly AI platforms, leveling the playing field for smaller agencies looking to compete.

But Z.ai (formerly Zhipu) is on the US Entity List due to its Beijing ties after OpenAI flagged its rapid progress. With this in mind, companies piloting open-source options should do so cautiously and consult with compliance teams before integrating.