This is Why They Banned Claude Fable 5

TL;DR

• 1

  Fable 5 feels like a new tier above Opus: Riley calls Anthropic's Mythos-class Claude Fable 5 the best model in the world for 99.9% of tasks, especially for long tool use, spatial reasoning, and visually faithful recreation.

• 2

  He rebuilt Lovable in two prompts and claims it worked better: Using screenshots plus Claude Fable 5, Daytona sandboxes, Convex, Vercel AI Gateway, and Vercel hosting, he made a Lovable clone called "Riable" and showed features like editable rich text and inline tables.

• 3

  Example-driven generation is absurdly good now: After uploading a McKinsey deck, Fable 5 produced a polished AI forecast report with matching chart styles, and Riley suggests asking Claude to fetch 20 downloadable report exemplars to improve outputs further.

• 4

  Games are a standout use case: Riley shows Fable building a simple Minecraft-style app, praises its performance on games, and points to demos from Pat Simmons and Balavvel Sudu as evidence that agent-built interactive software is getting much more visually sophisticated.

• 5

  The 'ban' is about hidden restrictions, not full access: He cites complaints that when users push into frontier LLM development topics like pretraining pipelines or accelerator design, Anthropic quietly swaps or constrains Fable 5 with prompt modification, steering vectors, PEFT, or a downgrade to Opus 4.8.

• 6

  The bigger thesis is the 'building block economy' for agents: Riley argues the next wave belongs to reusable components like Supabase, Convex, Neon, Daytona, plugins, skills, and SOPs, and says agentic payments will matter because agents still hit walls when a human must sign up, pay, or enter API keys.