Claude Code · verified November 16, 2026

Using Claude Fable 5.2 in Claude Code

Fable 5.2 is never the default on any plan — you select it. Here is every way to switch, what it costs, and how Anthropic says to prompt it.

The short answer

Run /model fable inside a session, or start one with claude --model fable. Both select Claude Fable 5.2 on Claude Code v2.1.321 or later. On Pro, Max, Ultra, Team and Enterprise plans Fable is selectable, but it is never the account default — and depending on your plan and seat tier its usage can bill to usage credits instead of your included limits.

Claude Code v2.1.321 or later · Anthropic's docs read on 2026-11-16

Three ways to switch, two variables behind them

Anthropic documents five places a model can be set. In practice you use the first three; the two environment variables cover scripts and fleet defaults.

  • /model fable — switch mid-session

    Typing /model fable switches immediately and saves the choice as the model field in your user settings, so later sessions start on Fable 5.2. Run /model with no argument to open the picker: Enter switches and saves, s switches for this session only.

  • claude --model fable — decide at launch

    The flag applies to the session you launch with it and is never saved. That is the supported way to run Fable in one terminal and Opus in another at the same time. claude --model claude-fable-5-2 selects the same model by its full ID.

  • model — make it the default

    Put "model": "fable" in ~/.claude/settings.json, or a project .claude/settings.json, and new sessions start on Fable 5.2. A model set in project or managed settings applies at every launch and outranks what /model saved; the startup header names the file that set it.

  • ANTHROPIC_MODEL and ANTHROPIC_DEFAULT_MODEL

    ANTHROPIC_MODEL sets the model for the session launched under it. ANTHROPIC_DEFAULT_MODEL (v2.1.236 or later) sets what new sessions start on, but only when no flag, environment variable, settings value or organization default has already chosen one. A third variable, ANTHROPIC_DEFAULT_FABLE_MODEL, changes what the fable alias itself resolves to.

  • Which one wins

    Anthropic's order, highest first: /model in the session, --model at startup, ANTHROPIC_MODEL, the model field in settings, then ANTHROPIC_DEFAULT_MODEL. The first two and ANTHROPIC_MODEL last only for the session; /model and the settings field persist.

The one line that sets it

"model": "fable"

The alias follows Anthropic's Fable default. Unless ANTHROPIC_DEFAULT_FABLE_MODEL is set, fable resolves to Fable 5.2 everywhere except Claude apps gateway sessions, where it resolves to Fable 5 — and a gateway that is not configured to serve claude-fable-5-2 rejects that model ID, so pin it with /model claude-fable-5-2 there. Use the full ID in this file if you want the version frozen.

Which method lasts how long

The same five steps as a table: what you set, and whether it survives the session.

Where you set itWhat it looks likeLasts
/model in the session/model fableThis session, and saved as your default
--model at startupclaude --model fableThis session only
ANTHROPIC_MODELANTHROPIC_MODEL=fableThis session only
model field in settings.json"model": "fable"Every new session
ANTHROPIC_DEFAULT_MODELANTHROPIC_DEFAULT_MODELNew sessions with nothing else set

Precedence runs top to bottom: a row further up beats one below it. A default your organization sets can override your saved choice, though --model and ANTHROPIC_MODEL still win.

Who can select Fable 5.2

Paid plans and the API can run it; zero-data-retention organizations cannot. No plan or provider makes it the default.

  • Pro, Max, Ultra, Team and Enterprise

    These are the plans Anthropic names when it explains how a saved claude-fable-5-1 pick moves to the fable alias, and Fable appears in their /model picker. Anthropic publishes no plan where Fable is the default: "No Fable model is the account-type default on any plan or provider."

  • Anthropic API and Console

    On the Anthropic API a Fable row appears in the /model picker, with its price on the row, unless availableModels or organization model restrictions exclude it. On Amazon Bedrock, Google Cloud and Microsoft Foundry the model has provider-specific IDs, so select it by that ID or pin ANTHROPIC_DEFAULT_FABLE_MODEL.

  • Not available under zero data retention

    Fable 5.2, 5.1 and 5 are Covered Models that require data retention by default. In an organization that cannot use them the row is missing or grayed out in the picker and the server rejects the request whatever the client is set to.

  • LLM gateways and the Claude apps gateway

    A gateway that is not configured to serve claude-fable-5-2 rejects requests for that model. To use Fable 5.2 through a gateway that does serve it, name the ID: /model claude-fable-5-2. Behind your own ANTHROPIC_BASE_URL, Claude Code keeps the 1M-token window it knows for the model.

  • Claude Code v2.1.321 or later

    Fable 5.2 needs Claude Code v2.1.321 or later; Fable 5.1 needed v2.1.257. An older version fails the request, because it does not know the model ID — run claude update, then start a new session.

What it costs, and when it leaves your plan limits

List rates per million tokens, then the plan mechanics that decide whether a Fable token spends your allowance or your usage credits.

Rate (USD / MTok)Claude Fable 5.2Claude Opus 5.5
Input$8$4
Output$40$20
Cache hit$0.2$0.2
Fast mode input$16$8
Fast mode output$80$40

Fable 5.2 list price is read from the model registry, verified against Anthropic's pricing docs on 2026-11-11; Opus 5.5 is the model Anthropic recommends starting on. Fast mode bills exactly 2x the base rate and does not combine with the Batch API. The full cache and batch breakdown is on /claude-fable-5-2-pricing.

  • "Requires usage credits" in the picker

    Depending on your plan and seat tier, Fable usage can bill to usage credits instead of drawing on your plan's included limits. When it does, the Fable row in /model says "Requires usage credits". Usage credits are billed at standard API rates, on top of your subscription.

  • Max and Ultra have a separate Fable weekly limit

    Anthropic's help centre states that Max and Ultra plans have an additional weekly limit for Fable, on top of the shared weekly limit and the five-hour session limit. Reaching one of them is what moves Fable traffic onto usage credits, when those are enabled.

  • The session and weekly windows are shared across models

    The five-hour session limit and the weekly limit are seat-based windows shared by every model, so switching to a smaller model with /model does not restore access. The message names the reset time.

  • Fast mode doubles the rate

    Fast mode keeps the same model and prioritises speed over cost. On Fable 5.2 it bills $16 in / $80 out per MTok and needs Claude Code v2.1.321 or later; fast mode persists across sessions on Fable 5.2 only if you turned it on while running Fable 5.2. Availability also depends on your organization's usage tier.

  • /usage and /usage-credits

    Run /usage for plan usage bars, per-model token counts and, once usage credits are on, a credits row with this month's spend against your limit. Run /usage-credits to open the billing settings for your plan, or to ask an admin for credits on Team and Enterprise.

  • Scripts bill without asking

    In non-interactive runs (-p) and in Agent SDK apps that do not show the consent prompt, Claude Code never asks before billing a Fable request to usage credits. If cost matters, cap the run with --max-budget-usd.

Effort, thinking and context

Fable 5.2 starts at medium effort, always thinks, and runs with a 1M-token window by default.

  • Five effort levels, saved per model

    Fable models support low, medium, high, xhigh and max. /effort sets a level and saves it for that model: Enter in the slider saves, s applies to this session only, and max is session-only unless it comes from CLAUDE_CODE_EFFORT_LEVEL. Claude Code stores the level under modelSettings.

  • medium is the default — raise it for bug hunts

    Fable 5.2 defaults to medium, which Anthropic describes as the fit for day-to-day work with a clear scope. high is for work where verification matters or edge cases are likely; max suits problems you want worked through without you, and Anthropic warns it can overthink. ultrathink in a prompt raises that one turn.

  • Thinking cannot be turned off

    Fable models always use adaptive reasoning. The thinking toggle and the /config row report that it cannot be turned off, and a saved alwaysThinkingEnabled: false or MAX_THINKING_TOKENS=0 has no effect. Effort, not the toggle, is how you spend less on thinking.

  • 1M tokens by default

    Fable 5.2 runs with the 1M-token window without an [1m] suffix, including on Bedrock, Google Cloud, Foundry and gateway sessions. Sessions auto-compact at roughly 967K tokens unless you set a smaller window with /autocompact.

  • Safety classifiers can move the session

    Fable models run with safety classifiers that most often flag cybersecurity and biology content. A flagged request re-runs on Opus 5 (biology) or Opus 4.8 (cybersecurity), and the session continues on that model until you run /model. Choose Ask each time in /config to decide per request.

Effort is calibrated per model: the same level name is not the same amount of thinking across models, so a level you liked on Opus 5.5 does not carry over unchanged.

How Anthropic says to prompt it

The four things Anthropic's Claude Code docs recommend for Fable, summarised rather than quoted in full.

  • Describe the outcome, not the steps

    Hand it the result you want and let it plan the path. If the work spans more than one session, set a goal so it keeps working toward that outcome.

  • Hand it the ambiguous problems

    Root-cause investigations, outage debugging and architecture decisions are where the extra investigation and verification pay for themselves. A ticket with a clear repro is where a cheaper model is fine.

  • Skip the verification reminders

    It verifies its own work with less prompting, so standing instructions to test or double-check usually just spend context. Keep the check itself — the test command, the lint target — and drop the nagging.

  • Size up your tasks

    Give it work you would normally break into pieces: it holds long sessions without losing the thread. That is also where the token bill lands, so watch /usage for the first few.

Three prompts to start from (our examples, not Anthropic's)

1. The checkout test suite fails intermittently in CI and passes locally. Find the root cause, fix it, and tell me what made it flaky — don't stop at the first hypothesis that explains the symptom.

2. Own this migration end to end: read the schema, propose the plan, get the build green, and summarise the changes and the risks I should review.

3. Goal: every page under /docs loads in under one second on a cold cache. Work through the causes, fix what you can verify, and leave me a list of what needs a decision.

The four items above are Anthropic's guidance; these prompts are our examples of what that looks like in a real session.

Sources

Every Claude Code claim on this page comes from one of these pages, read on 2026-11-16.

Claude Code ships often and the model plumbing moves with it: aliases, defaults and plan rules all changed several times in 2026. If /model does not offer Fable 5.2, run claude update first, then re-read the model configuration page above — that is the page this one follows.

Rolling Fable out to a team? Set model in managed settings instead of asking people to type /model, use availableModels with enforceAvailableModels to control the picker, and remember Fable is not permitted in zero-data-retention organizations.

Price it before you switch

Fable 5.2 is twice Opus 5.5 on input and output, and fast mode doubles that again. Run the numbers against your own token mix first.

Frequently asked questions