What Is Claude Fable 5?
Imagine an AI model with so much power that the US government had to literally step in and lock access for the entire world. That’s Claude Fable 5 — the most powerful model that has ever come out. People are creating games with it, building entire businesses, generating videos, and so much more.
Fable 5 is Anthropic’s next generation of intelligence, built for the hardest knowledge work and coding problems out there. It’s also the most expensive model on the market right now — roughly three times the price of GPT-5.5. But here’s what you get for that price: it’s the best model available for the most ambitious, long-running projects, thanks to its looping ability.
Run Fable 5 inside an agent harness like Claude Code, and it can work for days at a time — planning across multiple stages, delegating to sub-agents, and checking its own work continuously. It’s not just a tool that searches for an answer and hands it to you. It assesses the answer, decides if it’s good enough, and if it isn’t, it keeps working until it finds something better. Once that’s done, it moves on to the next problem and keeps going — for days at a stretch.
Fable 5 is Anthropic’s most capable model for ambitious coding projects, including large migrations, complex implementations, and multi-day autonomous sessions. On top of that, it understands diagrams, charts, tables, and dense documents nested inside files and PDFs — improving performance across document-heavy work in finance, legal, analytics, and architecture. This is exactly the kind of work companies are willing to pay millions of dollars for.
Why the Government Had to Step In
And because it’s this powerful, they had to make it safe — which is why access was blocked for 19 days.
Anthropic launched Claude Fable 5 on June 9, 2026. Just three days later, on June 12, access was suspended to comply with U.S. Department of Commerce export controls. The restriction stayed in place until June 30, and full access was restored on July 1 — a shutdown of roughly 19 days, exactly as widely reported.
What Makes Fable 5 Different
Just look at the physics. When given the same prompt as other top models — GPT-5.5, Claude Opus 4.8, and other leading LLMs — Fable 5 produced the most realistic physics simulation by far. The other models’ outputs always looked a little too fake. Fable 5’s collisions, object behavior, and real-world physics just look damn real.
The builder of Claude Code summed it up best: Fable 5 does in a day what used to take a team a whole month. Most people, though, are still using it wrong — which is exactly why prompting technique matters so much with this model (more on that below).
What People Are Building With It
Here’s what people have actually built with Fable 5:
A second brain for your business. Feed it your business structure, knowledge graphs, and goals, keep it alive with loops, and it becomes a research workflow that generates real insights — helping make your business, and everything you do, dramatically more productive.
A Subway Surfers clone, built in about an hour. From initial idea to game art, 3D models, and code — the entire thing was made with Fable 5. And it looks genuinely close to the real game.
A trading bot that won $288,000. Someone built an entire automated trading system using Fable 5 that found opportunities and traded on their behalf on Polymarket — and won $288,000 doing it. Fable 5 can genuinely be your trading partner.
A GTA 6 trailer, made entirely with Fable 5. The run went like this: minute 0, six reference frames dropped. Minute 9, a full shot list of 22 shots, timed to 2.1 seconds each. Minute 24, every scene generated, with 19 of 22 usable on the first pass. Minute 41, the sequence cut and matched to the trailer’s audio waveform. Minute 58, exported at 1080p, 91 seconds — from just four prompts, with Fable running the other 200 steps on its own: planning, generating, reviewing its own frames, and regenerating the bad ones. Compare that to the old pipeline — a VFX team, three weeks, $15,000, when a single fake trailer shot from a freelancer already costs $2,000. Rockstar spent 12 years and reportedly $2 billion making the real thing, with 6,000 people involved.
A full 3D map of San Francisco. Golden Gate Bridge, 2,600 buildings, ferries, fog — all regenerated and fully explorable. Work that would take a human days to rebuild from scratch, done in a few minutes.
Entire generated worlds using Three.js, with no external assets — worlds you can actually move through.
Portfolio websites and mobile app UIs that look like they cost thousands in design work — perfect fonts, perfect transitions, interactive stats, and views, all generated directly from a prompt.
The Smart Way to Use Fable 5: Orchestrator, Not Coder
Here’s the real trick when using Claude Fable 5: use Fable as the planner, and let Opus or Sonnet handle the actual code.
The setup people are using: Fable 5 with max reasoning as the orchestrator, Opus for deep reasoning on sub-agent tasks, and Sonnet for the mechanical work. Set up multiple agents this way inside Claude Code, and you can build any app you want — while saving a huge number of tokens along the way.
The logic is simple: the most valuable work isn’t writing the code. It’s deciding what to build, what it should look like, and how the different services should connect. That kind of thinking, reasoning, and logic is exactly what Fable 5 is best at.
How to Actually Prompt Fable 5
Here’s the part most people get wrong: prompting Fable 5 is not the same as prompting Claude Opus 4.8. This is a fundamentally different model with a different approach to building things, and Anthropic has published detailed guidance on how to prompt it effectively.
The key idea: Fable 5 is so powerful that the bottleneck becomes you, not the model. With Claude Fable, prompting is less about giving a perfect instruction up front, and more about helping the model discover hidden assumptions, constraints, preferences, and edge cases that might show up later. Stop treating it like a one-shot command, and start treating it like an iterative discovery process. The goal isn’t to write the perfect prompt — it’s to uncover your own blind spots. Fable isn’t just better at executing tasks; it’s better at discovering what the task actually is.
Instead of saying “build this feature,” try something like: “I’m new to this feature. I know the user goal, but I don’t know the architecture, the edge cases, or the steps in between. Help me identify what we’re missing before we move to implementation.”
The full prompting sequence looks like this:
- Blind-spot pass — find unknowns, unknowns, risks, and hidden constraints.
- Brainstorm — explore different directions.
- Prototype — build a first version of what you want.
- Interview — have it ask you one question at a time about anything ambiguous.
- References — point it to specific files and folders to use as context.
- Implementation plan — write the plan, leading with decisions, not just steps.
There’s also an effort setting to keep in mind — ranging from low all the way up to max. Higher effort means more thorough, better-reasoned responses, but it also takes longer and burns through your usage limits faster. Max effort uses excessive tokens and can result in long response times — save it for your hardest tasks only.
One more thing worth knowing: Anthropic has offered a window where users get up to 50% extra weekly usage on Fable 5 — after which extra usage gets billed at a higher rate. That’s part of why so many people are using Fable 5 purely for orchestration and planning, and letting cheaper models handle the actual code writing.
