LongPage is the first comprehensive dataset released by Pageshift-Entertainment in 2025 for training artificial intelligence models to write complete novels with complex reasoning capabilities.

This dataset currently contains samples from approximately 300 books, each ranging from 40,000 to over 600,000 words, covering novellas to epic novels. The data structure includes multi-level planning clues such as character archetypes, story arcs, world rules, and scene breakdown, as well as structural metadata such as dialogue density, pacing, and narrative focus. Each sample includes a prompt, a thought chain, and the complete novel structure (book). It supports training processes from cold-start supervised fine-tuning to reinforcement learning, making it suitable for training large-scale language models with hierarchical reasoning capabilities and improving the coherence and planning of long-form writing.