@trikcode Sometimes there will be false positives for coding tasks. They are working on improving the logic that determines what prompts are Fable restricted but until then the fallback will be Opus. Building a perfect F1 score LLM classifier with their scale and diversity is complex