Claude Fable 5 in Claude Code

Claude Fable 5 is free in Claude Code on Pro, Max, Team, and seat-based Enterprise plans until June 22, 2026. On June 23 it moves to usage credits, so the practical move is to run your biggest jobs while it is still included in your plan.

Fable 5 is Anthropic's first publicly available Mythos-class model, launched June 9, 2026. The model ID is claude-fable-5. It is state of the art on nearly every coding and agentic benchmark Anthropic tested, and the gap over Opus 4.8 grows the longer and more complex the task gets.

The banner in Claude Code says it plainly: "Fable 5 is here! Our newest model for complex, long-running work. Included in your plan limits until Jun 22, then switch to usage credits to continue."

This post is about getting the most out of that window. What to run before June 22, when Fable 5 actually beats Opus 4.8, and how to set effort so a long run finishes clean.

Quick Verdict

Reach for Fable 5 in Claude Code when the job is big and you can hand it off:

codebase-wide migrations across many files
large refactors you would normally schedule over weeks
overnight or async autonomous runs you walk away from
rebuilding a UI from screenshots alone
one-shotting a whole app that used to take dozens of prompts
long analytical or data jobs that run for a while

Stay on Opus 4.8 for everyday edits, quick Q&A, and cheap bulk automation. Fable 5 costs twice as much per token, and on short tasks its lead shrinks. The 2x price is worth it when the task is hard enough that fewer, better turns pay for the rate.

The Free Window, Exactly

Here is the timeline as Anthropic published it. The dates matter, so read them carefully.

When	What happens
June 9 to June 22	Fable 5 included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost
June 23	Fable 5 removed from those plans. Using it after that requires usage credits
After that	Anthropic aims to restore it as a standard subscription feature when capacity allows, and to communicate changes ahead of time

Two things sit outside the subscription clock. On the Claude API and consumption-based Enterprise plans, Fable 5 is fully available from day one and stays that way. The June 22 deadline is a subscription-plan thing, driven by capacity.

That capacity point is the why behind the staged rollout. Anthropic has been struggling to keep up with demand, and a model this popular adds real load. So it would rather give subscription users access sooner in a limited window than wait. The line "if capacity allows, we'll extend the included window" is a maybe, not a promise. Plan around June 22, not around the extension.

The community read on this is blunt and basically correct: the first taste is free. After June 23, continued use on a subscription means buying usage credits, which is metered consumption rather than your flat plan allowance.

The Free Window Playbook

Treat the next stretch as a chance to run the jobs you have been putting off because they were too big to babysit. Here is a concrete list.

Run the migration you have been avoiding. This is the headline use case. During early testing, Stripe used Fable 5 to run a codebase-wide migration across a 50-million-line Ruby codebase in a single day. The same work by hand would have taken a whole team over two months. If you have a deprecated internal API, a framework bump, or a language-version migration spread across a large repo, this is the job to queue.

Fan out a large refactor. Anything that touches many files at once plays to the model's strength. The longer and more complex the task, the larger Fable 5's lead, in Anthropic's own framing.

Kick off an overnight autonomous run. Fable 5 can work autonomously for longer than any previous Claude model. Hand off a feature or a bug sweep at the end of the day and let it run.

Rebuild a legacy UI from screenshots. Fable 5 is the new state of the art for vision tasks, and it can rebuild a web app's source code from screenshots alone. If all you have is screenshots of an old internal tool, this is a real path to a working rebuild.

One-shot an app that used to take a hundred prompts. Base44, a vibe-coding platform, said apps that took a hundred prompts a year ago, Fable 5 now one-shots. It is the model they reach for when a customer hits a wall.

Run a long analytical job. Hex reported Fable 5 was the first model to break 90% on its benchmark of complex, long-running analytical tasks, a 10-point jump over Opus.

Wire up file-based memory for a multi-session project. When Anthropic gave the model persistent file-based memory on a long task, it helped Fable 5 three times more than it helped Opus 4.8. If you maintain a notes file or memory store across sessions, Fable 5 gets disproportionate benefit.

Fable 5 vs Opus 4.8: The Decision Table

The price difference is the thing to keep in front of you. Fable 5 is $10 per million input tokens and $50 per million output, exactly double Opus 4.8 at $5 and $25. So the question is never "which is smarter," it is "is this task hard enough that the better model pays for itself."

Task type	Pick	Why
Codebase-wide migration	Fable 5	Months compressed into days; lead grows with complexity
Large multi-file refactor	Fable 5	Long-horizon work is where it pulls ahead
Overnight or async autonomous run	Fable 5	Works autonomously longer than any prior Claude
Screenshot-to-code rebuild	Fable 5	State-of-the-art vision, less scaffolding needed
One-shotting a whole app	Fable 5	Strong on vibe-coding and tool calling
Long analytical or data jobs	Fable 5	First to clear 90% on long-running analytical tasks
Everyday small edits	Opus 4.8	2x price is not justified for a quick change
Quick Q&A and lookups	Opus 4.8	Short tasks shrink Fable 5's lead
Cheap bulk automation	Opus 4.8	Cost across many sessions adds up fast
Cost-sensitive parallel work	Opus 4.8	Per-token rate dominates when tasks are easy

There is a speed angle that cuts against the price, too. On Anthropic's everyday spreadsheet suite, Fable 5 beat Opus 4.8 at every effort level and did it with fewer turns, finishing runs 25 to 30 percent faster. Fewer turns can mean fewer total tokens even at the higher rate. Measure on your own workloads before assuming the double rate means double the bill.

One more practical note for coders. Fable 5 ships with safeguards that route some prompts to Opus 4.8 instead, but they trigger in less than 5 percent of sessions overall, and almost never for ordinary app, feature, or migration work. The main thing that can trip them is security-adjacent territory. Even then you get an Opus 4.8 answer, not a refusal. We cover the full mechanism in the safeguards explainer.

What "Complex, Long-Running Work" Actually Means

The banner calls Fable 5 the model for "complex, long-running work." That phrase is doing real work, so here is the concrete version.

Complex means the task has many interdependent parts that you cannot fully script in advance. A migration where a discovery in one file changes how you handle forty others. A refactor where the right move only becomes clear after the model has read enough of the codebase. These are the tasks where Fable 5's lead is largest, because earlier models lose the thread partway through.

Long-running means the model holds focus across a lot of tokens and a lot of steps without a human checking each one. Anthropic says Fable 5 stays focused across millions of tokens in long-running tasks and improves its outputs using its own notes. Cursor put it as opening up "a class of long-horizon problems that were out of reach for earlier models."

The opposite of this is a quick, well-scoped edit. Change one function, fix one bug you have already located, rename a variable. Those are short tasks, and they are exactly where Opus 4.8 at half the price is the right call.

Effort Settings for Long Sessions

Effort is the dial that decides how much the model thinks and works per step. Fable 5 supports five levels: low, medium, high, xhigh, and max. For coding and agentic work, xhigh is the sweet spot, and it is the default in Claude Code for this model family.

The general rule is to use a minimum of high for anything intelligence-sensitive, and xhigh for serious coding and long-horizon runs. Effort matters more on Fable 5 than on any prior Opus model, so it is worth tuning rather than leaving on autopilot.

At the highest effort, the model reflects on and validates its own work before reporting back. Rakuten described this as the thing that makes highly autonomous operations possible, where "the extra thinking pays for itself." That self-checking is what makes an unsupervised overnight run trustworthy. The model is not just producing output, it is auditing it.

There is one habit that matters more than the effort number itself: give the full task spec up front, in one well-specified turn. Long-horizon models perform best with a clear goal stated at the start, rather than dribbled out across many follow-ups. In Claude Code, that maps to setting your goal at kickoff. State the task, the intent, and the constraints in the first message, then let it run at high or xhigh effort. Vague prompts spread across many turns tend to waste tokens and sometimes hurt the result.

If you want to control total spend on a long run, the API also offers task budgets, where you tell the model how many tokens it has for the whole agentic loop and it self-moderates against a running countdown. That is distinct from a hard output cap. It is a suggestion the model can see and pace itself against.

A Note on Cost After June 22

If you are on a subscription and want to keep using Fable 5 after the included window closes, you will be buying usage credits, which is metered consumption on top of your plan. Anthropic has not published a specific subscription credit price beyond pointing to the API rate of $10 and $50 per million tokens.

The honest planning advice is this. Fable 5 is a tool for big, valuable jobs, not your default editor. The economics only work when the task is hard enough that the better model earns its rate. For everything else, Opus 4.8 stays the workhorse at half the price. That split does not change after June 22. The free window just lets you find out, at no extra cost, which of your jobs are genuinely Fable-5-shaped.

Should You Switch Your Claude Code Default to Fable 5?

For most people, no, not as a blanket default. Make Fable 5 the model you deliberately reach for, the same way Opus 4.8 is the model you reach for over Sonnet.

Switch to Fable 5 for the run when:

you are about to start a migration or large refactor
you are handing off async work and walking away
you are rebuilding from screenshots or one-shotting an app
the task is long, complex, and the cost of a confident mistake is high

Stay on Opus 4.8 the rest of the time. The free window is the cheap way to build the instinct for which is which. Run a couple of your real big jobs on Fable 5 before June 22, watch how many turns it takes versus Opus 4.8, and let that decide where it earns a place in your workflow.

Frequently Asked Questions

Is Claude Fable 5 free in Claude Code?

It is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans from June 9 through June 22, 2026. On June 23 it leaves those plans and using it requires usage credits. On the Claude API and consumption-based Enterprise plans, it is fully available and billed from day one at $10 per million input tokens and $50 per million output.

When should I use Fable 5 instead of Opus 4.8?

Use Fable 5 for complex, long-running work: codebase-wide migrations, large refactors, overnight autonomous runs, screenshot-to-code rebuilds, and long analytical jobs. Stay on Opus 4.8 for everyday edits, quick Q&A, and cheap bulk automation, where its half-price rate wins and Fable 5's lead is small.

How much does Claude Fable 5 cost?

It is $10 per million input tokens and $50 per million output tokens, which is twice the price of Opus 4.8 at $5 and $25. The model ID is claude-fable-5. Because it often finishes in fewer turns, the total bill is not always double, so measure on your own workloads.

What effort level should I use with Fable 5 in Claude Code?

Use xhigh for serious coding and agentic work, which is the default for this model family in Claude Code, and a minimum of high for anything intelligence-sensitive. At the highest effort the model validates its own work, which is what makes long unsupervised runs trustworthy.

Why does Fable 5 sometimes answer from Opus 4.8?

Fable 5 ships with safeguards that route prompts in cybersecurity, biology and chemistry, or distillation to Opus 4.8 instead. They trigger in under 5 percent of sessions and almost never on ordinary coding work. You get an Opus 4.8 answer rather than a refusal, and you are told when it happens. See the safeguards explainer for the full picture.

What is the biggest job I can run during the free window?

Anything codebase-scale. Stripe used Fable 5 to run a migration across a 50-million-line Ruby codebase in a day, work that would have taken a team over two months by hand. Queue your largest migration or refactor before June 22 while it is still included in your plan.