01What Is Claude Fable 5.2 and How It Works

What is Claude Fable 5.2? It is the newest Fable model from Anthropic, built for the hardest problems, and it is live inside Claude, Claude Code and Cowork. In this episode, Julian Goldie explains what changed and tests it on three real business jobs.

Anthropic has a few families of Claude models:

  • Haiku is the small, quick one.
  • Sonnet is the everyday workhorse.
  • Opus is the big all-rounder most people use for serious work.
  • Fable sits above all of them, for jobs where other models give up halfway.

Version 5.2 came out on November 9th, 2026. At the same time, Anthropic released Mythos 5.2. They share the same brain underneath.

Fable is the version everybody can use, with extra safety checks built in. Mythos only goes to vetted security teams and science labs through special verification programs.

So if you run a business, Fable is the one for you. It is available on every plan and platform that includes Fable models, including the Claude app, Claude Code, Cowork and the developer platform. It is also on Amazon Web Services, Google Cloud and Microsoft Azure.

02What's New in Claude Fable 5.2

This update brings changes to speed, teamwork and safety. Here are the main ones Julian covers:

  • Faster answers. Fable 5.2 writes its answers up to 30% faster than Fable 5.1, with nothing to switch on.
  • Fast mode. This speeds up how quickly it writes by up to 2.5 times. It is available in Claude Code, on the Claude developer platform, on Google Cloud and on Microsoft Foundry.
  • The orchestrator role. Fable acts like a manager, breaks a big project into pieces and hands the simpler pieces to smaller models like Haiku.
  • The advisor role. Your everyday model can ask Fable 5.2 for advice when it hits something hard.
  • Fewer unneeded blocks. Anthropic says it has the same strength of safety checks but about 55% fewer unnecessary blocks in early testing.

Testers also found that Fable 5.2 communicates more clearly than the Fable models before it. It explains what it built and what it changed in plain words.

If a request does get blocked, it now hands it to another Claude model instead of just stopping. Anthropic says it was tested by outside groups, including the UK AI Security Institute and METR.

03How to Start Using Claude Fable 5.2

Julian lays out a simple way to start. You talk to Fable in plain English, the same way you would brief a person.

  1. Open Claude Code and pick Fable 5.2 as the model.
  2. Turn the effort up if the job is big.
  3. Describe what you want in plain English, in one message.
  4. Let Fable plan it, hand the easy parts to helper models and check everything before it comes back.
  5. Review what comes back and ask for changes. It updates while you watch.
  6. Save the message that worked so you can reuse it next time.

Then pick the next job and do one job at a time. Julian says the website, the dashboard and the launch in this episode each came from one message.

There is also good news if you are on a Claude plan. As part of the launch, Anthropic says that on Pro plans and standard Team seats, Fable can be used for up to half of your weekly usage. So you can lean on it for your biggest jobs.

04Three Real Business Jobs Julian Gave Fable 5.2

Julian tested Fable 5.2 on three jobs a small business might need.

Job one: a 3D website from one message. He asked for a site for a made-up wood-fired bakery brand called Ember and Oak, with a real 3D scene at the top, a menu, a timeline, reviews and a pre-order form. No images or templates. The oven is drawn with code and moves in real time, the menu cards slide in as you scroll, and the timeline draws itself.

You can keep talking to it afterwards, like asking it to change the colors to match your logo.

Job two: a sales dashboard. He gave it a file of numbers and asked it to work out what is really happening in the business. It built a dashboard where numbers count up, charts draw on, and a note sits right on the trend that matters most. At the bottom, a panel called What to do next lists three actions, each tied to a number.

Job three: a launch with helper agents. Fable planned a monthly subscription launch and handed three jobs to three Haiku helpers at once:

  • An Instagram carousel.
  • A landing page with a sign-up form.
  • Three short video ad scripts.

Fable then reviewed every file and fixed anything that did not match, so all three use the same name, colors and offer.

05Claude Fable 5.2 Test Scores and Results

Anthropic tested Fable 5.2 against its own models and models from other companies. Julian picks the tests that matter for real work and explains each one.

  • Remote Labor Index. Real paid online jobs like design, writing, data work and small builds. Fable 5.1 finished 17.9%. Fable 5.2 finished 41.3%. Opus 5.5 finished 21.3%.
  • Frontier Code 2.0. A coding test where Fable 5.2 scored 49.1%, ahead of Opus 5.5, OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon on Anthropic's chart.
  • OS World. Checks if AI can use a computer like a person, opening apps and clicking through menus. Fable 5.2 scored 89.3%.
  • Chartography. Reading charts and graphs. Fable 5.1 scored 44.8% and Fable 5.2 scored 73.4%.

Then there is the maths proof. Anthropic had Claude write out the proof of Fermat's Last Theorem in a strict computer language where a machine checks every step. The first attempt took 11 days.

This time, a team of about 1,000 agents worked together. With Opus 5.5 running the team, it got through about a quarter of the proof. With Fable 5.2 running it, they finished the whole thing in 30 hours.

Anthropic also says Fable 5.2 helped solve two open problems in black hole physics, and that over 90% of the changes to its own core code are now written by Claude with people guiding it.

06Effort Levels, Tips and Mistakes to Avoid

Julian says one setting decides how good your results are, and most people never touch it. It is called effort, which is how hard the model thinks before it answers.

Fable 5.2 has five effort levels:

  • Low
  • Medium
  • High
  • X high
  • Max

Here is what most people miss. In the Claude apps and in Claude Code, the default is medium. If you never change it, you are using Fable on medium.

Medium is fine for a quick email or a summary. But for a full website, a big piece of analysis or anything with many moving parts, turn it up. Anthropic's best scores mostly came from higher settings.

The coding score came from X high, the real job score came from high, and some science and design results came from max.

So the rule is simple: small job, leave it on medium; big job, turn the effort up.

Another tip is to use fast mode when you are building something live. Waiting is the slowest part of going back and forth. Shorter waits mean more rounds in the same hour, and more rounds mean a better result.

07The Old Way vs the New Way With Fable 5.2

Julian compares how big jobs usually get done with how they get done using Fable 5.2.

The old way: you have a big job, like a new website, a sales report or a product launch. You either do it yourself over a weekend or hire three or four people and lose the week chasing them. Every change means another email and another wait.

The pieces never quite match, and by the time it is done the moment has passed.

The new way:

  • You open Claude Code and pick Fable 5.2.
  • You turn the effort up for the big job.
  • You describe what you want in plain English.
  • Fable plans it, hands easy parts to helper models and checks everything.
  • You review it and ask for changes, and it updates while you watch.

This is why Fable does well on big jobs. A real website has hundreds of moving parts, like the 3D scene, scroll effects, layout, form and colors. Older models get one part right and break another.

Fable keeps the whole plan in its head and checks its own work as it goes.

08Who Claude Fable 5.2 Is For and Why It Matters

Fable is built for the hardest problems, but Julian points out that a website, a dashboard and a launch are hard problems for a business owner too. If you have a shop, an agency or any business with a spreadsheet you never have time to study, these are the jobs to hand to Fable.

Julian also answers a few doubts:

  • Is it only for scientists and engineers? No. You talk to it in plain English, like briefing a person.
  • Do I still need Opus or Sonnet? Yes, for quick questions and everyday writing. Anthropic suggests starting there for most everyday work. Reach for Fable when the job is big or needs to be right the first time.
  • Can I use both? Yes. With the advisor role, your everyday model asks Fable only when it gets stuck, like a junior staff member knocking on the expert's door.

It matters because the jump in this update is large. It finishes more than double the real jobs of the last version, reads charts far better, and can run a team of helper models for you. Julian also mentions that the website, dashboard and launch systems from this episode are inside the AI Profit Boardroom.