AI5 min read

Anthropic's Fable 5.2: what changes and when to use it

Fable 5.2 came out on November 9, 2026: 20% cheaper, up to 30% faster output and fewer blocked requests. What changes for you, in plain language.

Fable 5.2 is the most capable model Anthropic makes available to everyone, released on November 9, 2026. Compared with 5.1 it is 20% cheaper, writes up to 30% faster, scores higher on almost every test and blocks about half of the harmless requests it used to stop. For everyday work, though, Anthropic itself still recommends Opus 5.5.

Same depth of reasoning, faster answers

Fable 5.2 at a glance

Fable 5.1Fable 5.2
ReleasedSeptember 1, 2026November 9, 2026
Input price$10 per million tokens$8 per million tokens
Output price$50 per million tokens$40 per million tokens
Output speedbaselineup to 30% faster
How much text it reads at once1 million tokens1 million tokens
Longest possible answer128 thousand tokens128 thousand tokens

A token is a piece of a word: a million tokens is roughly half a million words. Prices apply to API use, meaning inside a piece of software. If you use Claude through a subscription, what you pay does not change.

For developers, the identifier is claude-fable-5-2.

What actually improves

It costs less

Input drops from $10 to $8 per million tokens and output from $50 to $40. Reading from the cache (text the model has already read and does not need to process again) drops from $0.25 to $0.20.

Price per million output tokens
  • Fable 5$50
  • Fable 5.1$50
  • Fable 5.2$40

Anthropic list prices, November 2026.

It writes faster, and there is an even faster mode

Answers arrive up to 30% sooner than with 5.1. There is also an optional fast mode: up to 2.5 times faster, at double the price ($16 input, $80 output). It is available in Claude Code, on Anthropic's developer platform, on Google Cloud and on Microsoft Foundry.

It does better on coding and real-work tests

A benchmark is the same exam given to every model, so they can be compared. Anthropic published many: these two are the most useful for seeing the difference.

FrontierCode 2.0: autonomous coding on real projects
  • Fable 5.132.4%
  • Opus 5.545.5%
  • Fable 5.249.1%

Anthropic data, November 9, 2026, maximum effort.

Remote Labor Index: real freelance jobs delivered to an acceptable standard
  • Fable 5.117.9%
  • Opus 5.521.3%
  • Fable 5.241.3%

Anthropic data, November 9, 2026.

The biggest jump is on the second one: more than twice as many jobs completed as 5.1. It also gets much better at reading difficult charts and images (from 44.8% to 73.4% on a chart-reading test).

These are Anthropic's numbers, measured by Anthropic. Useful for the direction. The test that matters is still your project.

It blocks fewer harmless requests

Fable has safety filters on two sensitive topics: cybersecurity and biology. With 5.1 it sometimes blocked ordinary questions too. In Anthropic's early tests 5.2 blocks about 55% less (about 60% less on cybersecurity, 40% less on biology).

When a request is still stopped, you are not left without an answer: a less powerful model replies instead (Opus 4.8 for cybersecurity, Opus 5 for biology).

It uses less on hard tasks

In the Claude apps and in Claude Code, effort (how much the model reasons before answering) starts at medium. At that level 5.2 finishes demanding tasks with fewer tokens than 5.1: less waiting and, through the API, less spend.

Fable 5.2 and Mythos 5.2: what is the difference

Mythos 5.2 came out alongside Fable 5.2. It is the same base model, with fewer filters on cybersecurity and biology. That is why Anthropic only gives it to verified organizations, through its cybersecurity and biomedical research programs. Enterprise customers can also find it inside Claude Security, the tool that scans code for vulnerabilities.

For a normal business the choice is simple: Fable 5.2.

And against the competition? I compared Fable 5.2 with GPT-6, Gemini 4 Carbon, Grok, the Chinese models and the open-weight ones in a separate article: who wins, what it really costs and where your data goes.

When to use it, and when not to

The most interesting part of the announcement, for me, is that Anthropic does not present it as the model to use all the time. For daily work it recommends Opus 5.5, which costs half as much. It suggests Fable 5.2 in two roles:

  • advisor: other models do the work and only ask it when they hit a hard problem;
  • orchestrator: it splits a big job into pieces and hands the simple parts to smaller models.

That is how I use it too. Opus for everyday work. Fable when there is a bug nobody can explain, a long migration, an analysis that has to be right the first time.

One example Anthropic cites: a materials research company had Fable 5.2 optimize a scientific computing program. Across 13 test calculations, run time went from 3.3 hours to under 4 minutes.

Two things to know before choosing it:

  • on Pro and on standard Team seats, Fable can use at most half of the weekly limits;
  • data is retained for 30 days for safety monitoring. Enterprise customers with dedicated safeguards can use it with zero retention.

My take

Fable 5.2 is a real step forward, especially on long, autonomous work. But the right message is the one Anthropic itself gives: you do not need the most powerful model switched on all the time. You need to know when to bring it out.

If you want to understand whether a model like this makes sense for an automation or a custom business tool you have in mind, write to me. I start from the problem, not from the model.

If you want to know what came before, I also wrote about Fable 5 and Fable 5.1.

Sources: Anthropic announcement, November 9, 2026, Fable page and developer documentation.
FAQ

Common questions, answered simply

Can't find your question? Message me on WhatsApp

Pro, Max, Ultra, Team and Enterprise, plus the API for developers. On Pro and on standard Team seats, Fable can use at most half of the weekly limits: the rest goes to the other models.

Let's talk

Want to apply it to your business?

Tell me about your case in a 30-minute discovery call: I'll tell you if and how I can help.

Book a discovery callAll articles

30 minute call · free · no commitment

Keep reading