Summary
Chris Raroque tests Fable 5, Anthropic’s newest coding model that was available for roughly 48 hours before being pulled by the US government. During that window, Raroque used the model extensively on a personal backlog of ambitious features that both Opus 4.8 (ultra code setting) and GPT 5.5 Extra High had previously failed to solve.
The video walks through three real shipping tasks: a multi-column drag-and-drop refactor for the iPad version of his daily planning app Ellie (involving ten-plus edge cases), a cross-list Kanban drag system for Ellie’s web version, and a creative emoji-rain animation for his calorie tracker Amy. Fable 5 reportedly solved the first two in one and two prompts respectively — tasks where competing models gave up after ten or more iterations. Raroque also validates his results using Greptile, an AI code review service, noting that all Fable 5 pull requests scored four or five out of five on the first pass, an unusually clean result.
The video concludes with a cost-benefit discussion about API pricing and whether the capability jump justifies paying full price when the model becomes available again. For iOS and web developers working on complex, multi-state interaction problems, this is a useful real-world data point on where frontier coding models currently stand relative to one another.
📺 Source: Chris Raroque · Published June 17, 2026
🏷️ Format: Review
