Claude Fable 5: Anthropic's Most Capable Model Explained

Claude Fable 5 is Anthropic's most capable widely-released model, built for the very hardest reasoning and longest agentic runs. It has a 1M-token context window, up to 128K output tokens, and premium pricing of $10 per million input and $50 per million output tokens.

Pricing, context limits, and model IDs verified 31 July 2026 against Anthropic's models overview.

That price tag is the whole story in miniature. Fable 5 sits above the everyday flagship, and you reach for it only when a problem genuinely earns the extra cost. For most work, a cheaper model is the smarter call. This guide explains what the model is, where it fits, exactly what it costs, and how to decide between it and the tier just below it.

What Claude Fable 5 is

Model ID claude-fable-5, the top of the lineup by raw capability and tuned for demanding multi-step reasoning and long agentic runs. It is widely released, so any developer with API access can call it without an invitation.

Claude Fable 5 (model ID claude-fable-5) is the top of Anthropic's lineup by raw capability. It is tuned for demanding, multi-step reasoning and for agentic tasks that run for a long time without losing the thread — deep research passes, sprawling refactors, complex analysis that has to hold many moving parts in mind at once.

It is a widely-released model, so any developer with API access can call it. That matters, because Anthropic also ships an access-restricted sibling with the same brains (more on that below). If you want the most capable Claude you can actually use today without an invite, Fable 5 is it.

One practical note: the model has somewhat different API behavior than the Opus family. If you are porting a prompt or an agent harness over from Opus, plan to test rather than assume a drop-in swap. Treat a migration as a small project, not a config change.

Where the model sits in the lineup

Top rung of a four-model ladder: Fable 5 at $10/$50, Claude Opus 5 at $5/$25 as the agentic coding flagship, Sonnet 5 at $3/$15 for balanced work, and Haiku 4.5 at $1/$5 for high volume.

Anthropic's mainstream lineup is a ladder, and Fable 5 stands on the top rung above the everyday flagship. Below it, Claude Opus 5 (claude-opus-5) is the coding and agentic flagship and the right default for most hard work, with extended thinking on by default and the same 1M-token context; the older Claude Opus 4.8 remains available as a legacy option at the identical $5/$25 price. Claude Sonnet 5 is the balanced workhorse, and Claude Haiku 4.5 is the fast, cheap option for high-volume tasks.

Here is how the tiers compare on the numbers that decide most projects:

Read the ladder top to bottom and the design intent is clear: Fable 5 is not a replacement for Opus. It is an escalation path for the handful of tasks where Opus leaves capability on the table and the extra spend is worth it. If you want the full family tree, our Claude models explained overview lays out every current tier.

Pricing, context, and output specs

$10 per million input tokens, $50 per million output, a 1,000,000-token context window where 1M is both the default and the maximum, and up to 128K output tokens per response. Roughly double what Opus 5 costs.

The headline figures are simple to state and important to internalize, because Fable 5 is the premium frontier tier and its cost reflects that.

• Input: $10 per million tokens
• Output: $50 per million tokens
• Context window: 1,000,000 tokens — 1M is both the default and the maximum
• Max output: 128K tokens per response

Compare that to Opus 5 at $5 input and $25 output and the math is blunt: Fable 5 costs roughly double. On a token-heavy agentic job that reads large repositories and writes long outputs, that difference is not rounding error — it can be the line between a $40 run and an $80 one. Our Anthropic Claude API pricing breakdown shows how those per-token rates translate into real monthly bills once caching and volume enter the picture.

The 1M-token context is generous and, usefully, there is no smaller default to trip over — you get the full window every time. For example, you can drop an entire mid-size codebase, a long research corpus, or a stack of transcripts into a single request and let the model reason across all of it. If you want the general mechanics of how Claude handles long inputs, see our guide to the Claude context window.

When to choose Claude Fable 5 vs Opus 5

Escalate only on three signals: reasoning deep enough that a small quality gain matters, agentic runs long enough that drift gets expensive, or stakes where being wrong costs far more than the token bill. Otherwise stay on Opus 5 or Sonnet.

This is the decision that actually matters, so be honest with yourself about it. Most users, most of the time, should stay on Opus or Sonnet. Claude Fable 5 earns its price only when a task is genuinely at the ceiling of what the tier below can do.

Reach for Fable 5 when:

1. The reasoning is unusually deep. Multi-layered analysis, tricky proofs, or planning problems where a small quality gain has outsized value.
2. The agentic run is very long. Tasks that chain dozens of steps and need to stay coherent across all of them, where drift is expensive.
3. The cost of being wrong is high. When a better answer is worth paying double for because a mistake costs far more than the token bill.

Stay on Opus 5 — or drop to Sonnet — when the work is ordinary hard rather than extraordinary hard. For agentic coding specifically, Opus 5 is the flagship and usually the better pick; our best Claude model for coding guide walks through why the coding-tuned flagship, not the pricier tier, wins most day-to-day. If you are still weighing the middle of the ladder, the Claude Sonnet vs Opus comparison covers that trade-off in depth.

A simple rule of thumb: start on the cheapest model that plausibly works, measure, and escalate only if the output quality actually falls short. Escalating to Fable 5 by default is how you burn budget without a matching gain.

We route models by task every day building this site, and being honest about our own ladder is the best argument we can make for this advice: Sonnet handles the bulk of the work, Opus 5 is where we escalate when something is genuinely hard, and the top rung is somewhere we step only for a problem that has actually pushed the flagship to its limit. That happens far less often than it feels like it should when you first meet Fable 5. Our discipline is to make the jump a deliberate decision with a reason attached, never the default we reach for out of a vague sense that "more capable" must mean "better here."

The Mythos 5 and Project Glasswing relationship

Claude Mythos 5 (claude-mythos-5, code-named Project Glasswing) shares the same capabilities, the same $10/$50 pricing, and the same API surface. The only difference is access — Mythos is invitation-restricted, Fable is the public door to identical capability.

Claude Fable 5 has a twin. Claude Mythos 5 (model ID claude-mythos-5, internally code-named "Project Glasswing") shares the same capabilities, the same $10/$50 pricing, and the same API surface. The one real difference is access: Mythos 5 is access-restricted, the successor to the invitation-only Claude Mythos preview, while Fable 5 is the widely-released version anyone with API access can use.

Think of them as one model behind two doors. If you have a Mythos invitation, you already have Fable-5-class capability; if you do not, Fable 5 gives you the same power through the public door. There is no capability penalty for using Fable 5 instead of Mythos 5 — the choice is purely about which door is open to you.

How to access the model

Set the model ID to claude-fable-5 in a standard API request. You need API access and a payment method on file; there is no waitlist. Test prompts rather than copying an Opus harness verbatim, since API behavior differs somewhat.

Getting started is the same as any other Claude model. You call it by its model ID, claude-fable-5, through the standard API. There is no separate signup or waitlist for the widely-released model.

To use it in practice:

• Set the model ID to claude-fable-5 in your API request.
• Confirm your account has API access and a payment method on file, since this is a paid, premium tier.
• Budget for the premium rate — at $50 per million output tokens, output-heavy jobs add up fast, so cap your token usage where you can.
• Test your prompts rather than copying an Opus setup verbatim, because the API behavior differs somewhat from the Opus family.

The official pricing page lists the current rates for every tier side by side, the models overview documents the model IDs, context windows, and output limits you will need when wiring up a request, and the broader Claude developer docs cover the API mechanics for calling any model in the lineup.

As of this writing, Claude Fable 5 is the newest and most capable widely-released Claude, sitting one rung above Claude Opus 5, which remains the flagship most teams should default to. That is the current state of the lineup: a premium ceiling for the hardest problems, and a strong, cheaper flagship for nearly everything else.