This article details an experiment comparing the cost-effectiveness and output quality of Kimi K2.7 Code versus Claude Fable 5 for generating landing pages. The authors found that while standard prompts produced generic results, integrating a custom MCP server to provide visual design context significantly improved layout and hierarchy in the lower-cost open-source model.
Key points
Open-source models like Kimi K2.7 Code can produce competitive design outputs compared to frontier models at a fraction of the cost.
Text-only prompts often result in generic AI aesthetics; multimodal inputs are required for professional design quality.
Using a custom Model Context Protocol (MCP) server to provide visual references improves typography, hierarchy, and composition.
Iterative design workflows benefit significantly from combining low-cost model inference with targeted visual context.