Claude Fable 5. The New Rules of Prompting. 12 Moves That Unlock Full Potential

Fable 5 is not a better version of the Claude you already know how to use.

The prompting habits that worked on Sonnet and Opus, step-by-step instructions, asking it to show its reasoning, micromanaging the approach, actively get in the way with Fable 5. It's built differently. It thinks differently. And it needs to be prompted differently.

These are the twelve moves that actually unlock what it can do.

The shift in how you prompt

1. Stop giving it steps - give it goals

Older Claude models needed hand-holding. You'd break the task into steps, tell it exactly how to approach each one, and guide it through the process. That worked because those models needed the structure. Fable 5 doesn't - and giving it that structure actually limits what it can do.

Fable 5 is built to figure out the how on its own. Give it a destination, not a route. The more you micromanage the path, the less it can use its own judgment to find a better one.

❌ Giving it steps:

First, research the topic. Then write an outline. 
Then write the introduction. Then write each section 
one by one. Finally, write the conclusion.

✅ Giving it a goal:

I need a comprehensive article on [topic] aimed at 
[audience]. The goal is [what you want the reader 
to walk away with]. Make it [tone]. 
Here's everything I know about it: [context].

The second prompt gives Fable 5 room to work. The first one just slows it down.

2. Don't ask it to show its reasoning

With older models, asking Claude to "think step by step" or "show your chain of thought" was one of the most reliable ways to get better output. It forced the model to slow down and work through the problem before answering. That advice is everywhere - and with Fable 5, it's wrong.

Fable 5 does adaptive thinking on its own. It already reasons through problems before responding. Asking it to narrate that process doesn't improve the output - it interrupts it. In some cases it can trigger a refusal entirely, because you're asking it to expose internal reasoning it's not designed to surface.

If you want to understand what Fable 5 did and why, ask it after the fact:

That's good. Can you summarize what approach you took 
and why you structured it that way?

That gets you the insight without getting in the way of the thinking.

3. Set the effort level before you start

Most people don't know this setting exists. Fable 5 has three effort levels - and which one you use changes how wide the model looks at your problem, not how smart it gets. Getting this wrong means either burning through tokens on a task that didn't need it, or getting a shallow output on something that deserved more.

|Effort   |  Best for | Example  |
| --- | --- | --- |
|  Max |  Complex creative work, deep reasoning, high-stakes output |Long-form writing, research synthesis, hard strategic problems   |
| Medium  |  Most standard tasks | Drafting, analysis, coding, research  |
| Low  | Fast, simple outputs  |  Summaries, quick edits, formatting |

The default is medium. For anything that matters, set it explicitly - don't let Fable 5 guess what level of effort the task deserves.

[Set effort to max]
I need you to think carefully about this. 
The task is: [your task]

4. Give it more context than you think it needs

Fable 5 has a 1 million token context window. That's roughly 750,000 words about 10 full-length novels worth of text in a single conversation. Most people are still prompting it like it's a model with a 32k window, rationing context because they're used to older constraints that no longer exist.

Stop editing down. Stop summarizing when you could paste the original. Stop giving Claude the highlights when it can handle the whole thing.

Here's what most people hold back - and shouldn't:

Fable 5 doesn't get confused by more information. It gets sharper. The people getting the most out of it aren't writing tighter prompts, they're giving it everything and letting it find what matters on its own.

Where Fable 5 actually shines

5. Long tasks it can run on its own for hours

This is the capability most people haven't tested yet. Fable 5 isn't just better at answering question, it's built to run. Give it a goal with enough context and it can plan across stages, execute each one, check its own work, and keep going without you in the loop. Not for minutes. For hours.

The instinct is still to break big tasks into small prompts and check in after each one. That made sense with older models - they'd drift, lose context, or need correction every few steps. Fable 5 holds the thread. It remembers what it decided three hours ago and why. It catches its own mistakes before surfacing the output.

What this actually unlocks:

I'm going to give you a large task and I need you to 
run it end to end without checking in unless you hit 
something genuinely ambiguous. Here's the goal: [goal]
Here's everything you need: [full context]
Flag anything that requires a decision. Otherwise, finish it.

6. Creative writing - but specifics are everything

Fable 5 is the best creative writing model Anthropic has ever released. The problem is most people prompt it like they're filling out a request for "write me a short story about loss" - and get back something technically competent and completely forgettable. The model isn't the limitation. The prompt is.

Generic input gets generic output. Always. What separates a Fable 5 creative output that stops you mid-read from one you close after two paragraphs is specificity in the setup, the characters, the tone, the constraint, the thing you actually want to feel at the end.

Don't describe the style. Don't say "make it emotional" or "make it feel cinematic." Show it what you mean:

Write a short story about a father and his adult daughter 
who haven't spoken in three years meeting at a funeral. 
The tone should feel like this paragraph: [paste example]
The ending should leave the reader uncertain about whether 
they reconcile - don't resolve it cleanly.
The voice should be sparse. Short sentences. No metaphors.

The more specific the constraint, the more Fable 5 has to work with. Specificity isn't limiting - it's the brief. And Fable 5 executes briefs better than any model before it.

7. Complex reasoning problems older models couldn't hold

There's a category of problem that previous Claude models could start but not finish. Problems where the reasoning chain is long, the variables interact in non-obvious ways, and staying consistent across the whole thing requires holding more in mind than the model could manage. Fable 5 is the first publicly available Claude model where that category almost disappears.

It doesn't just answer hard questions, it sustains hard thinking. It tracks assumptions across a long chain of reasoning, catches when an early decision contradicts a later one, and corrects course without losing the thread. The kind of problem that used to require you to break it into five separate prompts and stitch the outputs together manually, Fable 5 handles in one run.

Anthropic demonstrated this on launch day - Fable 5 built a full simulation of the solar system, deriving each planet's orbital motion from first principles physics, and used it to predict a solar eclipse. No shortcuts, no lookup tables. Just reasoning from the ground up.

That's not a party trick. That's what the model does with hard problems when you give it room to work.

8. Agentic work - let it plan, delegate, and check itself

Most people use Fable 5 as a very powerful chatbot. That's the lowest-ceiling version of what it can do. Fable 5 was built for agentic work tasks where it doesn't just answer a question but runs a full workflow: planning the stages, executing each one, delegating to sub-agents where needed, and checking its own output before surfacing the result.

The shift is treating Fable 5 less like a tool you operate and more like someone you brief. You give it the goal, the constraints, and the context. It figures out the plan. It runs it. It comes back when there's a genuine decision to make - not every five minutes for reassurance.

This changes what's possible. Not just faster execution of things you already do, but entire workflows you previously couldn't run without a team.

I need you to work autonomously on this. 
Here's the goal: [goal]
Here are the constraints: [constraints]
Here's all the context you need: [context]

Plan the approach first, then execute it stage by stage. 
Only stop and check in if you hit something genuinely 
ambiguous that requires my input. Otherwise run it to completion.

If you want to go deeper on everything Fable 5 can do:

The Code with Claude Tokyo 2026 keynote covers the full picture - what Fable 5 was built for, how the team thinks about agentic work, and where this is all heading. It's 42 minutes and worth every one of them.

Everything after this makes more sense once you've seen it.

Mistakes people carry over from older models

9. Micromanaging the approach

The habits that made you good at prompting older Claude models are the ones that hold you back with Fable 5. Specifying the format before it starts. Telling it which angle to take. Breaking the task into sub-tasks and handing them over one by one. Checking in after every step to make sure it's on track.

With Sonnet and Opus, that level of control helped. The models benefited from structure because they needed it to stay on course. Fable 5 doesn't. It has its own judgment about how to approach a problem and that judgment is usually better than the step-by-step plan you'd hand it.

Micromanaging Fable 5 is like hiring the best person you've ever worked with and then standing over their shoulder telling them how to hold the pen. The control doesn't improve the output. It limits it.

Give it the goal. Give it the context. Get out of the way.

10. Starting fresh when it can hold the whole project in context

With a 1 million token context window, Fable 5 can hold an entire project in a single conversation - the brief, the drafts, the feedback, the decisions, the direction changes, all of it. Most people still open a new chat every session out of habit. Every time they do, Fable 5 loses everything it knew about the work.

This isn't just inconvenient. It's actively expensive. You spend the first part of every session rebuilding context that was already there. Fable 5 makes assumptions it wouldn't have made if it remembered the last conversation. You correct the same things twice.

Here's what resets every time you open a new chat:

For any project that spans more than one session - stay in the conversation. The longer Fable 5 runs with your work, the better it knows it.

11. Using it for simple tasks - Sonnet is faster and a fraction of the cost

Fable 5 costs $50 per million output tokens. Sonnet costs a fraction of that. For the tasks that make up most of people's daily Claude usage - summarizing, drafting emails, quick edits, formatting, simple research, Sonnet gets you 95% of the result at a fraction of the price and often faster.

The Code with Claude Tokyo 2026 keynote covers the full picture what Fable 5 was built for, how the team thinks about agentic work, and where this is all heading. It's 42 minutes and worth every one of them.problems, agentic runs that need sustained judgment. Using it for everything because it's the most powerful model available is like taking a Formula 1 car to the supermarket. Impressive. Expensive. Wrong tool for the job.

A simple rule:

| Task  |  Right model |
| --- | --- |
|  Emails, summaries, quick drafts |  Sonnet |
|  Research, analysis, standard writing |  Sonnet |
| Complex reasoning, long projects  |  Fable 5 |
|  Creative work that needs to be exceptional |  Fable 5 |
|  Agentic runs, multi-day tasks | Fable 5  |

Save Fable 5 for the work that actually needs it. You'll get better results and spend less doing it

12. Use voice samples for creative work - don't describe the style, show it

Telling Fable 5 to write "in a conversational tone" or "like a journalist" or "punchy and direct" gets you its best guess at what those words mean. That guess is fine. It's not yours.

The fix is simple and most people never try it. Instead of describing the voice you want, paste an example of it. A paragraph from a writer you admire. A section from your own best work. A piece that sounds exactly like what you're going for. Fable 5 reads the sample and writes to match it, not the label, the actual thing.

Write [your task] in this voice:

[Paste 2-3 paragraphs of the style you want]

Match the sentence length, the rhythm, the level of 
formality, and the way it handles transitions. 
Don't describe the style back to me - just write in it.

The difference between "write like Hemingway" and pasting three paragraphs of Hemingway is not small. One gives Fable 5 a label. The other gives it a target.

Fable 5 is a different model. Not just more powerful - genuinely different in how it works and what it responds to. The prompting habits that got you good results before will get you average results here.

The twelve moves in this article aren't tricks. They're the adjustment - from prompting a model that needed guidance to working with one that doesn't. Make that shift and the gap between what you were getting and what you start getting is not small.