A significant bug in Anthropic's "Fable" model causes it to drop approximately 26% of user messages before they are received, leading to silent data loss. This issue, distinct from other Anthropic models like Opus and Sonnet, appears to occur when Fable processes text between two internal thinking steps, particularly before a tool call. Users have reported instances of Fable asking if they reviewed lists that were never provided or presenting unanswerable prompts, which are now understood to be consequences of this data loss. Despite being reported over a month ago, there is no indication of a fix, and the problem persists as Fable approaches API pricing. AI
IMPACT This bug could erode user trust in Anthropic's models and impact adoption if not quickly resolved.
RANK_REASON Bug report on a specific model feature impacting user experience.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
