4 exercises to test the new model and stop using it at half power.
Idriss Laouali Abdou
Co-founder Karatou Academy · AI Capacity Builder
News · AISeptember 3, 2026·7 min read
On September 1, 2026, Anthropic launched Fable 5.1, billed as the world's most advanced model for coding and knowledge work. A Mythos 5.1 version, restricted to vetted cybersecurity and life science professionals, ships alongside it.
Maybe you skimmed the announcement. "Another model." That's a mistake, and here's why.
The real story isn't one more model. It's the gap, now huge, between what these tools can already do and what most people still ask of them. These tools have improved enormously over the past few months. If you haven't tested one recently, you have no idea what they can do today.
So instead of walking you through the specs, here are 4 concrete exercises. Do at least one this week.
What actually changed (beyond the marketing)
Three things deserve your attention.
01
It goes the distance
It handles long tasks without losing the thread. It keeps track, reprioritizes, and picks up where it left off. That's what was missing for agents to become reliable.
02
It costs less
Anthropic reports around 25% savings on typical workloads, up to 45% on heavily automated tasks.
03
It has effort levels
Low, medium, high, and beyond. At low or medium effort, 5.1 already matches the previous generation, for a fraction of the price. Knowing how to dial it in becomes a real skill.
One telling example, from Anthropic: the model tracked down the cause of a crash a team of engineers had been chasing for years. It disassembled an external library to isolate the problem. Keep the idea, not just the anecdote: these models are no longer simple writing assistants, they're analysis partners.
The 4 exercises to do this week
1
Have it audit a project you already have
Take something real: an app, a website, a contract, a business plan, a deck. Ask three simple questions: what's broken, what's missing, what would you change. Don't expect it to be right about everything, look for the blind spots. The ones you've carried for months because you're too close to see them.
Prompt to copy
You are a senior consultant. Here is my project: [paste your text here: website copy, plan, contract, slides, business plan].
Review it with a critical, honest eye. Give me:
1. The 3 most serious problems, most urgent first.
2. What's missing that should be there.
3. The 3 concrete changes you'd make first, and why.
Be direct, no flattery. If anything is unclear, ask me before answering.
2
Build an agent
An agent is a system that gets a task done without you behind every step. Start tiny: a daily email summary, a watch on your industry, a document sort. Until now agents crashed too much to be serious. Not anymore. In two years, handing a whole task to an agent will be as ordinary as using a spreadsheet today.
Prompt to copy
I want to build a simple agent that [describe your repetitive task, e.g. every morning, summarize my unread emails and surface the top 3 to handle first].
Walk me through how to set it up, step by step, with the simplest tools possible. Assume I'm not a developer. Give me the first step to do today, then wait for me to confirm before moving on.
3
Give it the problem you're avoiding
You have a hard problem you've been putting off for months because it's too big or too fuzzy. Describe it in detail: the context, the constraints, what you've already tried, what's really blocking you. Then read the answer. The quality of the analysis is surprising, precisely because the model doesn't carry your emotional history with the problem. It won't solve it for you, but it often makes the first step visible.
Prompt to copy
I want to think through a hard problem I've been avoiding.
The problem: [describe it in detail].
My context: [your situation, your constraints].
What I've already tried: [...].
What's really blocking me: [...].
Don't hand me a ready-made solution. Start by restating the real problem as you understand it, ask me 3 questions to dig deeper, then propose 3 different angles of attack.
4
Learn to dial the effort levels
The exercise almost nobody does, and the one that changes your bill. Take one task, run it at low, then medium, then high effort. Compare the quality and the cost. Many tasks don't need the maximum setting. Save high effort for the problems that deserve it. Dialing it in means you stop wasting.
Prompt to copy
[Take a real task, e.g. draft a 3-hour AI training plan for beginner managers.]
Do this task as well as you can.
(Tip: run this exact prompt once at low or medium effort, then again at high effort. Compare the quality and the cost. You'll quickly see which tasks don't need the maximum setting.)
What this changes for your training
At Karatou Academy, our stance doesn't move: understand, don't just use.
A professional who can audit a project with AI, hand a whole task to an agent, dial the effort and check the work, is worth ten who know how to write a nice prompt. The difference isn't technical. It's in the way you work.
The AI world we taught six months ago no longer exists. Every release like Fable 5.1 reshuffles the deck. That's exactly why continuous training isn't a luxury, it's a necessity.
The question is no longer "will AI change my work?". It's: "which of these 4 exercises am I doing before Friday?" Do one today. Then tell me what it gave you.
Frequently asked questions
It's the same model. Fable 5.1 is the public version. Mythos 5.1 has reduced safeguards and stays restricted to vetted cybersecurity and life science professionals.
Idriss Laouali Abdou
Co-founder of Karatou Academy, AI Capacity Builder. Idriss supports organizations and professionals in the practical adoption of AI.
AI can hallucinate. Always check what it generates.
Want to actually use these tools on YOUR projects?
At Karatou Academy, we don't teach prompting. We teach a new way of working. Join the channel so you don't miss the next guides.
