Chapter generation with a local Ollama model takes hours in BookyAI but minutes in Ollama itself. Why?

Thinking models (Qwen 3, Gemma 4, DeepSeek R1 and similar) reason before every answer, and at local speeds that reasoning can take longer than the writing. When a chapter hit the output limit, BookyAI asked the model to continue, and the model reasoned again before each continuation, so one chapter could mean four rounds of reasoning. Update to version 1.9.4 or later, where thinking is off by default for Ollama; a chapter then takes about as long as in Ollama directly. You can turn thinking back on in Settings if you want it.