Ten models walk into a bedtime story (danwitt.substack.com)

A new AI tool called Storybook Forge has been developed to create personalized bedtime stories for children, combining multiple language models to enhance narrative quality. The tool automates story generation by first planning a narrative, allowing an LLM (large language model) to draft it, using a critic model for refining, and finally employing an illustrator to complete the book. Initially using local models, the creator discovered that cloud-based models outperformed them, spurred on by a unique micro-benchmarking method that evaluates both cost and quality of various AI authors. This experimentation has led to insights on model performance, including comparisons of ten different AI models tasked with writing the same story. Significantly, the benchmark not only serves as a guiding tool for future AI applications in creative writing but also reveals new contenders in the generative AI space. For instance, recent tests showed that the kimi-k3 model offered a remarkable balance of cost and quality, performing comparably to the leading model, gpt-5.6-sol, while being substantially more affordable. Additionally, the latest critic model, gpt-5.6-luna, demonstrated effectiveness in replicating human panel rankings for story quality, further emphasizing the potential of this AI-driven approach for creating nuanced, engaging narratives tailored specifically for children. This toolkit could pave the way for advanced storytelling applications in education and entertainment, highlighting the evolving intersection of AI and creative writing.