THE Claude Fable 5 model has reportedly experienced a decline in intelligence as users noted low self-reported reasoning effort scores despite high configuration settings. Although developers set the reasoning to high, the model only acknowledged a score of 10 out of 100, indicating minimal cognitive exertion. Screenshots shared by developers illustrated these concerns, and further testing showed scores of 40 with the extreme 'xHigh' effort setting.
This issue has raised dissatisfaction in the developer community, particularly as there were no updates about these changes in the changelogs from Anthropic, leading to suspicions of undocumented server-side adjustments. An official response clarified that the numerical effort mappings do not correlate directly to performance and may adjust according to the complexity of user prompts.