Seongwoo Choi / Journal

Claude Fable 5.2 launch: new prices and how to switch

At a glanceAnthropic released Claude Fable 5.2 on November 9, 2026 and cut API prices by 20%. What changed, who can use it, the cost in practice, and how to turn it on.

Not testedA summary of Anthropic's published prices and documentation. The model was not tested.

In this article 7 sections
  1. 1. What was announced
  2. 2. The prices, old and new
  3. 3. What that means for a real task
  4. 4. Who can use it
  5. 5. How to switch in Claude Code
  6. 6. What to know before relying on it
  7. 7. Should you switch?

On November 9, 2026, Anthropic released Claude Fable 5.2 and lowered the price of its top generally available model by a fifth. For anyone running coding agents or long automated jobs, the price is the news: the same kind of work now costs 20 percent less. This explainer covers what was announced, what it costs in practice, who can use it, and how to switch.

1. What was announced

  • A new version. Claude Fable 5.2 replaces Fable 5.1 as Anthropic's most capable generally available model.
  • A price cut. Input falls from $10 to $8 per million tokens, and output from $50 to $40.
  • A sibling with limited access. Claude Mythos 5.2 is listed at the same prices, with limited availability.
  • A faster option. A fast mode offers up to 2.5 times the output speed at twice the price.

2. The prices, old and new

Fable 5.1$50 per million output tokens

$50

Fable 5.2$40 per million output tokens

$40

Opus 5.5, for comparison$20 per million output tokens

$20
Output price per million tokens, from Anthropic's pricing table. Bars are drawn to scale.
Per million tokensFable 5.1Fable 5.2Change
Input$10$8−20%
Output$50$40−20%
Cache read$0.25$0.20−20%
Cache write, 5 minutes$12.50$10−20%
Cache write, 1 hour$20$16−20%

Two modifiers from the pricing documentation are worth knowing:

  • Fast mode: $16 input and $80 output per million tokens, for up to 2.5 times the output speed.
  • US-only inference: 1.1 times the standard price on every token category.

For scale, the same table lists Claude Opus 5.5 at $4 input and $20 output. Fable 5.2 costs exactly twice as much, so it is the model for the hardest work, not for everything.

3. What that means for a real task

Agents reread their context on every step, so most of their input is served from the cache. Take a long coding task that uses 2 million input tokens, 1.5 million of them from the cache, and writes 100,000 tokens:

TokensAt 5.1 pricesAt 5.2 prices
Fresh input500,000$5.00$4.00
Cached input1,500,000$0.38$0.30
Output100,000$5.00$4.00
Total$10.38$8.30

The token counts are an invented example; the arithmetic is exact. Notice where the money goes: 100,000 output tokens cost as much as 500,000 fresh input tokens, and 1.5 million cached tokens cost thirty cents. Keeping the start of your prompts stable, so the cache can be used, saves more than any model switch.

4. Who can use it

You areAccess
An individual subscriberPro, Max and Ultra plans
An organizationTeam and Enterprise plans
A developer using the APIThe Claude Platform, and through Amazon Web Services, Google Cloud and Microsoft Foundry
On the free planNot listed
Check how your plan bills itThe Claude Code documentation says that, depending on your plan and seat tier, Fable usage can be billed to usage credits instead of your plan's included limits, and that Claude Code can ask for your consent before continuing. Read that prompt before you agree.

5. How to switch in Claude Code

Fable is not the default model on any plan. You select it yourself:

/model fable

That is typed inside a session. To start a session on it from your shell:

claude --model fable
DetailWhat the documentation says
Minimum versionClaude Code 2.1.321 or later. Check with claude --version.
Is the choice remembered?Yes. /model saves it as the default for new sessions.
This session onlyOpen the picker with /model and press s on the model's row.
You had 5.1 savedOn paid plans the saved setting moves to the fable alias, which now means 5.2.
SubagentsThose that inherit the main model follow the switch.

On Windows and macOS the commands are the same. If /model fable reports that the model is not supported, update Claude Code first.

6. What to know before relying on it

  • Some requests are rerouted. Anthropic says requests flagged by its safety classifiers, most often in cybersecurity and biology, fall back to another model automatically. The product page names Opus 4.8 for cybersecurity and Opus 5 for biology. If you do security work, expect this, and look at the verification program the page describes.
  • Data is retained for safety monitoring. The product page states a 30-day retention by default, and describes an enterprise option compatible with zero retention.
  • Mythos is a separate track. It shares the price list but has limited availability.
  • Cheaper is not the same as cheap. At twice the price of Opus 5.5, routine edits and questions are better sent to a smaller model.

7. Should you switch?

Your workSuggestion
You already use Fable 5.1Switch. The price is lower and the alias moves you anyway.
Long, multi-step tasks that smaller models abandon or get wrongTry it on one such task and compare the result and the bill.
Everyday edits, questions and small fixesStay on your default model.
Batch jobs where speed does not matterStandard mode; fast mode doubles the price.
You are sensitive to cost surprisesSet a spending limit first, and watch /usage for a week.

The fair test is your own task. Run the same job on your current model and on Fable 5.2, and compare what you got and what it cost. For habits that cut the cost on any model, see AI coding agent tips.

We also covered this release on our channel, AI2Now: Anthropic cut its top model price by 20%.

Sources

This article reflects public information available when it was written. Versions, defaults and prices can change; check the linked official documentation.

Prices, availability and behaviour are from Anthropic's product page, pricing documentation and Claude Code documentation as read on November 12, 2026. This is a summary of published information, not a test of the model's quality: no benchmark was run for this article, and Anthropic's own benchmark claims are not repeated here. Prices change; check the official pricing page before budgeting. Written for this site with AI assistance, from the sources above. For errors or additional information, please request a correction. Editorial policy