FREE RESOURCE · FROM THE "LAUNCH" REEL

Claude Fable 5.2: 12 things it can do, and how to try them

By Hammad Yousuf · 12 November 2026

Anthropic released Claude Fable 5.2 on 9 November 2026. This page lists the twelve claims from the reel with what Anthropic's launch post actually says, then the steps to try the ones you can try today.

Read this first

Every number below is Anthropic's own, from its launch post. Nothing has been independently tested yet.
Fable 5.2 is not on the Free plan. Pro pays for it with usage credits at API rates, apart from a launch week that ends 16 November 2026. Max, Ultra and Team premium seats get it inside half of their weekly limit.
Mythos 5.2 is the same model with fewer safeguards, for vetted organisations only.
Anthropic's own advice in the same post: start with Opus 5.5 for most coding and knowledge work, and move up to Fable for the biggest, most open-ended jobs.

The twelve, with the source wording

Better vision. Chartography (charts, no tools): 73.4%, up from 44.8% for Fable 5.1. Unity: "rebuilding an experience from a visual reference".
Robot control. "Controlling physical robots, even for robotics tasks on which it has never explicitly been trained." Anthropic's video shows a puzzle cube and a bottle cap. Ordinary users cannot connect a robot today.
Faster by default. "Up to 30% faster than Claude Fable 5.1." API price 8 USD in and 40 USD out per million tokens, down from 10 and 50.
Advisor. Haiku, Sonnet or Opus does the work and consults Fable on the hard decisions.
Orchestrator. Fable plans and delegates simpler tasks to smaller models.
Many sub-agents. In one experiment it "coordinated over 1,000 subagents" for eight days and produced a prototype operating system that boots on a laptop. The system was not released.
Long unattended work. An Uber staff engineer ran "a 10 to 12 hour security audit" that came back with a write-up, fix pull requests and a policy proposal.
Computer use. OSWorld 2.1 (offline set, partial): 89.3%. In the apps, computer use is a beta for Pro, Max and Ultra.
Business and finance work. Box: it flagged that salary deposits and a bank statement's balance did not match in a loan file.
Self-correction. Asked to design three electronic devices, "all three devices worked without needing any electrical modifications".
Honesty. Cognition: it "says it doesn't know things instead of hallucinating" and said its last estimate was wrong.
Fewer wrong refusals. "About 55% fewer blocks" than Fable 5.1, with safeguards Anthropic describes as just as robust.

Try it in Claude Code

Install:

macOS, Linux, WSL

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell

irm https://claude.ai/install.ps1 | iex

Run claude in a project folder and sign in in the browser. A paid plan or a Console account is required.
terminal
```
claude
```
Check the version: claude --version must show 2.1.321 or later. If not, run claude update.
terminal · check
```
claude --version
```
terminal · update
```
claude update
```
Use Fable 5.2: /model fable. On Pro the picker shows "Requires usage credits" and asks for consent once.
claude code
```
/model fable
```
Advisor pattern (point 4): stay on /model sonnet or /model opus, then /advisor fable. Turn it off with /advisor off.
claude code · on
```
/advisor fable
```
claude code · off
```
/advisor off
```
Many agents (point 6): type ultracode: before a task, or say "use a workflow". Watch with /workflows. On Pro, switch on "Dynamic workflows" in /config.
task prefix
```
ultracode:
```
watch the run
```
/workflows
```
Computer use (point 8): in the desktop app, Settings > Capabilities > "Enable computer use". It is a beta and asks for approval per app.

Try it on the API

Model id: claude-fable-5-2
Price: 8 USD per million input tokens, 40 USD per million output tokens

model id

claude-fable-5-2

Prompts to test the claims yourself

1 · Chart reading test

prompt 1 of 3

Here is a chart image. Read every value on it into a table, then tell me which values you are least sure about and why.

2 · Numbers that do not add up

prompt 2 of 3

Here are a bank statement and three salary slips. Check whether the deposits match the slips and the closing balance. List every mismatch with the page and line.

3 · Honesty check

prompt 3 of 3

Before you answer: if you do not know, say "I don't know" and tell me what you would need to find out.

Official links

Sources checked

Launch on 9 Nov 2026; up to 30% faster than Fable 5.1; Chartography 73.4% vs 44.8%; OSWorld 2.1 89.3%; over 1,000 subagents for eight days; about 55% fewer blocks; advisor and orchestrator roles; robot control; three devices worked; Unity, Uber, Box and Cognition quotes; availability on AWS, Google Cloud and Azure. Source ↗
API price 8 USD in and 40 USD out per million tokens, down from 10 and 50. Source ↗
Essay 'We Must Pace the Frontier', September 2026. Source ↗
Advisor setup in Claude Code (/advisor). Source ↗
Computer use is a beta in the Claude apps. Source ↗

About the video: the reel that sent you here is presented by an AI avatar of Hammad Yousuf.