Skip to content

Cheat sheet

Claude Fable 5.2: what's new, benchmarks and best use cases

What changed in Claude Fable 5.2, how it scores against Opus 5.5 and GPT-6, the jobs it's best at and which plans include it.

On this page

Claude Fable 5.2 is Anthropic's most powerful AI model that anyone can use. It came out on 9 November 2026. It's built for big, long jobs, like building a whole app or taking a research project all the way to a finished report.

For everyday chats and writing, Opus 5.5 is still the better pick. Anthropic itself recommends starting there, and Opus is faster and uses less of your plan.

Fable 5.2 at a glance

SpecClaude Fable 5.2
Made byAnthropic, the company behind Claude
Released9 November 2026
Best forLong, hard jobs: building apps, deep research, checking numbers
Where to use itThe Claude apps, Claude Code, Claude Cowork and the API
PlansPro, Max, Ultra and Team (not the free plan)
API price$8 per million tokens in, $40 per million out
Memory per chat1 million tokens, about 550,000 words
SpeedUp to 30% faster than Fable 5.1

A token is a small chunk of a word. AI companies charge by the token when you use their models through the API (the way developers plug AI into their own apps).

What's new in Fable 5.2

Fable 5.2 replaces Fable 5.1, which came out earlier in 2026.

  • It's faster. Answers come back up to 30% faster. In Claude Code, fast mode makes it up to 2.5 times faster again, at double the price.
  • It's cheaper. API prices dropped 20%, from $10 and $50 to $8 and $40 per million tokens.
  • It's better at coding. It scored 49.1% on FrontierCode 2.0, a test of hard coding projects. Fable 5.1 scored 32.4%.
  • It reads images much better. On Chartography, a test of reading charts, it jumped from 44.8% to 73.4%.
  • It finishes long jobs. On the Remote Labor Index, a test made of real freelance projects, it more than doubled Fable 5.1's score.
  • It blocks fewer normal requests. Fable has safety filters for hacking and biology. They now wrongly block about 55% fewer requests than Fable 5.1's did.
  • It explains itself more clearly. Early testers said it's easier to follow what it did and why.

Fable 5.2 vs Opus 5.5: which Claude should you use?

Claude comes in four models. Fable is the most powerful and the most expensive.

ModelBest forAPI price (in / out)
Haiku 5.5Quick, simple tasks at scaleFrom $0.10 / $0.50
Sonnet 5.5Fast everyday work$2 / $10
Opus 5.5Your daily driver for writing, research and coding$4 / $20
Fable 5.2Big, long, hard jobs$8 / $40

Anthropic's own advice is to start with Opus 5.5 for most work. Move up to Fable when Opus can't get a job done.

That's how I use them too. Opus 5.5 is still my daily driver. I switch to Fable 5.2 when I'm building something big in Claude Code or handing off a long job.

The difference shows up on bigger jobs. One early tester found Fable hard to tell apart from Opus on ordinary tasks, and only saw the gap on more ambitious ones. The benchmarks below tell the same story.

Fable 5.2 benchmarks

Benchmarks are standard tests that AI companies use to compare models. These scores come from Anthropic's launch post, with each model at a high effort setting.

BenchmarkWhat it testsFable 5.2Fable 5.1Opus 5.5GPT-6 Astra
Remote Labor IndexReal freelance projects41.3%17.9%21.3%20.8%
FrontierCode 2.0Hard coding projects49.1%32.4%45.5%42.8%
SWE-bench ProFixing real software bugs93.0%81.2%89.9%Not tested
Terminal-Bench 4.0Working in the terminal67.6%55.8%66.4%57.9%
OSWorld 2.1Using a computer by itself89.3%83.5%87.2%79.5%
Humanity's Last ExamExpert questions, with tools74.2%65.6%67.7%57.2%
ChartographyReading charts73.4%44.8%64.4%71.0%
AutomationBenchBusiness workflows51.4%31.4%42.5%41.4%
AA-Briefcase (Elo)Knowledge work, head to head1785166918001569

ARC Prize, which runs its puzzle tests independently, checked Fable 5.2 itself. It scored 93.8% on ARC-AGI-2, a set of visual puzzles built to test reasoning on problems the model hasn't seen before.

What the numbers mean

The biggest jump is on real projects. The Remote Labor Index uses real paid freelance jobs, and Fable 5.2 scored nearly twice what Opus 5.5 and GPT-6 did. If you want to hand off a whole job, that's the number to look at.

Fable can also be cheaper than it looks. It can work at different effort levels. At a lower level, it matched Opus 5.5's best coding score for about half the cost per task. It matched Opus 5.5's best on Humanity's Last Exam for under a tenth of the cost.

It doesn't win everything. Opus 5.5 still edges it on head-to-head knowledge work. Google's Gemini 4 Carbon beats it on AutomationBench, with 53.9%. And on Terminal-Bench, the gap to Opus is small enough to be luck.

Most of these are Anthropic's own tests, run on its own setup. Independent leaderboards like Artificial Analysis and LMArena hadn't scored Fable 5.2 yet when I wrote this.

The best use cases for Fable 5.2

Use Fable 5.2 when a job has lots of steps and you want it finished without babysitting. These are the jobs where it's worth it for a small business.

1. Build a whole website or app in one go

Fable 5.2 keeps working until the job is done. Uber's team said it "works until the job is actually complete instead of stopping at a plausible checkpoint." That makes it great for building a full website, a booking tool or a client portal in Claude Code from one detailed brief.

Prompt
Build me a booking website for my dog grooming business. It needs a home page, a services page with prices, a booking form that emails me each request, and a contact page. Use a friendly, modern design in green and white. Keep going until every page works, then give me the steps to put it online.

2. Take a research project all the way to a finished report

Anthropic built Fable 5.2 for research that ends in a finished document, spreadsheet or deck. Give it a big question and a list of what you want back.

Prompt
Research the top 10 competitors to my Melbourne pilates studio. For each one, find their prices, class types, reviews and how they market on Instagram. Put it all in a spreadsheet, then write a 2-page report on where the gaps are and 5 ways I could stand out.

3. Check your numbers for mistakes

Testers keep praising how well it spots numbers that don't add up. Box said it caught salary deposits that didn't match a bank statement. Another tester said it noticed regional numbers that didn't sum to the total. Use it on your budget, your P&L or a quote before it goes out.

Prompt
Here's my profit and loss for the last 12 months [upload the file]. Check every total adds up. Flag anything that looks wrong or unusual, then tell me the 3 biggest costs I could cut.

4. Read screenshots, charts and dashboards

Its image reading jumped more than any other area. Screenshot your Meta Ads Manager, your Shopify dashboard or your Google Analytics and ask what's going on.

Prompt
Here are screenshots of my Meta ads dashboard for the last 30 days. What's working, what's wasting money and what should I change this week? Explain it like I'm new to ads.

5. Hand off long jobs to an agent

In Claude Cowork, Claude's agent mode, Fable 5.2 can run a long list of tasks on its own. Think sorting a folder of 200 receipts into an expense sheet, or going through every page of your website and listing what's out of date.

When not to use Fable 5.2

Stick with Opus 5.5 (or Sonnet 5.5) for:

  • Everyday writing. Captions, emails and ad copy don't need it. Dan Shipper at Every gave its writing a B to B-plus and said it's slow for everyday writing.
  • Quick questions. Fable thinks hard before almost every reply. That's slow and wasteful for simple asks.
  • Anything when you're near your limit. Fable uses up your plan much faster than other models.

It also needs a human eye. Ethan Mollick, a Wharton professor who tested it early, found its first attempts were accurate but lacked taste. He said it "did not notice any of these problems itself." It also tends to over-build, adding more than you asked for. Tell it exactly how long or simple you want the result.

How to use Fable 5.2

  1. Open the Claude app or claude.ai on a paid plan
  2. Open the model picker in the chat box
  3. Pick Fable 5.2

In Claude Code, type /model and pick Fable 5.2. It's now the default Fable model there.

You can also set how hard it thinks. The Claude apps start on medium effort. Anthropic says most of the gain comes by high, so only go higher for your hardest jobs.

Which plans include it

PlanFable 5.2 access
FreeNot included
Pro and Team standardIncluded up to half your weekly limit until 16 November 2026, plus one free reset. After that, you pay through usage credits
Max, Ultra and Team premiumIncluded up to half your weekly limit, plus 1 to 3 free resets depending on how long you've subscribed
API$8 per million tokens in, $40 per million out

A reset gives back the Fable usage you've spent that week. Find it under Settings > Usage > Limit resets. The Max and Ultra resets last until 8 March 2027, so you can save them for a big project.

What is Claude Mythos 5.2?

Mythos 5.2 is the same model as Fable 5.2, with fewer safety limits. Anthropic only gives it to vetted cybersecurity and biology organisations. Fable is the version everyone else gets.

You won't need Mythos unless you run security testing or lab research.

Questions

What is Claude Fable 5.2?

Claude Fable 5.2 is Anthropic's most powerful public AI model, released on 9 November 2026. It's built for long, hard jobs like coding whole apps, deep research and checking complex numbers.

Is Fable 5.2 better than Opus 5.5?

On big, long jobs, yes. It wins most of Anthropic's benchmarks and nearly doubles Opus 5.5 on real freelance projects. For everyday writing and quick questions, Opus 5.5 is just as good, faster and uses less of your plan.

Is Claude Fable 5.2 free?

No. It's not on the free plan. Pro and Team standard seats get it included until 16 November 2026, then through paid usage credits. Max, Ultra and Team premium seats include it up to half the weekly limit.

How much does Fable 5.2 cost?

In the Claude app, it comes with Max (from $100 a month), or with Pro ($20 a month) through usage credits. On the API, it costs $8 per million tokens in and $40 per million out, 20% less than Fable 5.1.

What's the difference between Fable 5.2 and Fable 5.1?

Fable 5.2 is up to 30% faster and 20% cheaper on the API. It's better at coding, reading charts and finishing long jobs, and its safety filters wrongly block about 55% fewer requests.

Is Fable 5.2 good for writing?

It's fine, but it's not the best use of it. Early reviewers rated its writing a B to B-plus and found it slow. Use Opus 5.5 for captions, emails and sales pages.

Follow me for more AI workflows

New guides and tools go out as short videos first. The write-ups land here.

From the library

  • Claude

    AI Assistants & LLMs

    4.6

    Anthropic's chat assistant, at its best on careful writing and long documents.

    Freemium
  • Claude Code

    Coding & Dev Tools

    4.4

    An AI agent that works inside your project: it reads, writes and runs the code for you.

    Paid
  • Claude Cowork

    AI Browsers & Agents

    4.1

    Claude's agent mode: hand it a whole task across your files, apps and browser.

    Paid

More free resources