Woofun AI reports that blogger Leo alleges the DeepSeek V4 Pro API silently routes complex programming requests to Claude Fable 5, potentially to collect outputs for model distillation. Tests using OpenCode to generate a 3D game revealed results with high similarity to Fable 5 and sudden reasoning shifts; however, introducing network security and biology questions caused quality to decline and knowledge scope to regress to DeepSeek's baseline. This behavior mirrors Anthropic’s routing of sensitive queries to Fable 5, leading testers to suspect initial routing to Fable 5 triggers a classifier fallback to DeepSeek. Existing evidence only confirms abnormal API behavior, failing to verify actual responders or whether outputs entered training data, as prompt mixing or undisclosed updates could also explain the discrepancies.