Testing my OpenRouter or Grok key says "Empty response" even though the key is valid. Why?

This was a bug, fixed in version 1.9.5. The Test button sends a tiny request with room for only a few words of reply. Reasoning models (such as DeepSeek R1, Claude with thinking, Qwen 3, Grok mini) think before they answer, and that thinking used up the whole allowance, so the reply contained reasoning but no text and BookyAI reported "Empty response". The key was fine and the request had gone through. Since 1.9.5 the test uses a proper allowance and a reasoning-only reply counts as a pass, shown as "Key works". Book generation was never affected because chapters have a far larger allowance.