How are Chinese models so strong with so little investment?
A tiny Chinese lab is beating Google's best model with a fraction of the budget.
Deep Dive
The open-source Chinese model GLM 5 is reportedly beating Google's Gemini 3 Pro in many benchmarks, a model considered state-of-the-art just three months ago. This raises a massive question about the ROI of Western AI investment, as Chinese labs achieve ~95% of the capability with just 1-10% of the capital and offer much cheaper inference costs, all while being open-source and hostable on private infrastructure.
Why It Matters
This could completely undermine the business case for the hundreds of billions being spent on AI infrastructure by Western tech giants.