Wow, the era has changed.
Honestly, I did a double-take at my screen.
Right now, if you open X, it's overflowing with talk about "Claude Fable 5."
Apparently, it's incredible.
But many people probably think, "I've heard the name, but I don't really get it," and just scroll past.
I understand that feeling all too well.
Actually, I was once someone who had given up on AI.
That's why I created this article.
First, I'll say the most important thing
Before the main topic, just one incredibly important point.
Fable 5 was just released on July 2nd.
And this Fable 5 can be used without additional cost on Pro and Max plans until July 7th.
Here are the facts as stated by the official source:
・Until July 7th: Available without additional cost
・From July 8th: It will temporarily be excluded from plan coverage, requiring the purchase of usage credits
・Afterward: Planned to return to standard plan inclusion once server capacity is ready
In other words, this week is the only time you can touch the strongest model for free.
It's not that you'll lose money if you don't do it now.
It's just a waste to miss this one week where you can experience the strongest model for free without knowing it.
That's all.
So, now is the only time to read this.
Let's get started together.
What is Claude Fable 5 anyway?
I won't use difficult terms. You only need to remember one thing:
"For the first time, something even better than the previous strongest model has been released to the public."
That's it.
Until now, Claude had a staircase of performance:
・Haiku (Fast)
・Sonnet (Balanced)
・Opus (Strongest)
Fable 5 is even higher than Opus, which was at the top. Anthropic calls this new class the "Mythos class."
Moreover, this model wasn't released immediately after completion. The reason is amazing.
When they tested the pre-release version, it found weaknesses in thousands of systems on its own. If it can find weaknesses, it could also be used for hacking if misused.
So, Anthropic first gave it to "defensive" organizations to fix those weaknesses. Now that work is done, it's finally public. That's the level of power currently in your hands.
To those who think "It's probably just overhyped"
"AI evolution is just noise every time."
I understand that skeptical feeling. I was on that side for a long time.
When I first touched AI, I concluded: "It talks well because people say it's great, but I can't trust it with real work."
It would lose track of the conversation during long tasks. It couldn't write code that actually worked until the end. I dismissed it as "not as good as the hype" and left it alone.
But this time, I stopped in my tracks. The reason wasn't just impressions; I was hit with actual measurements.
Here is the verification from the payment infrastructure giant, Stripe.
Migrating 50 million lines of code. A task that would take a human two months, Fable 5 finished in one day.
This isn't my opinion. These are numbers officially released by a world-class engineering team that supports global payments.
Look at the results of the "SWE-bench Pro" test, which measures coding ability:
・Claude Fable 5: 80.3%
・Previous strongest Opus 4.8: 69.2%
・Other company's latest model GPT-5.5: 58.6%
・Other company's latest model Gemini 3.1 Pro: 54.2%
It's 11 points ahead of the previous strongest and over 20 points ahead of other companies. You are free to doubt it, but if so, you should just verify it yourself this week while it's free.
What this article promises
Don't worry. This article is not for engineers.
・I will rephrase all technical terms on the spot
・All you have to do is "copy and paste"
・I won't give you choices; I'll lead you down a single path
"It seems difficult for me."
I accept that resistance head-on. It's okay. If you move your hands according to this article, you will definitely reach the same screen.
If you fall into these categories, please read to the end:
・People who hear AI names but don't really understand them
・People who were disappointed by AI once before and left it alone
・People who have ignored it thinking "It has nothing to do with my job"
If any of this applies, this article is for you.
You won't understand the greatness with short questions
I'll be honest upfront. You won't understand Fable 5's greatness with ordinary questions. For short questions like "What's the weather tomorrow?", there isn't much difference from previous AIs.
The official statement is clear: "The longer and more complex the task, the larger the gap between Fable 5 and other models."
The difference appears when you hand over a "heavy job you were hesitant to do because it was tedious" in its entirety.
This is the core of this article. So, I'll have you experience that difference in one shot.
The moment I did a double-take, now for you too
I want you to do one thing: "Reproduce a working web app from a single screenshot."
That's all.
What I actually did was very simple. I attached just one screenshot of the Yahoo! Weather forecast screen. Then I just pasted: "Reproduce the appearance and functionality as faithfully as possible."
I didn't add any explanation. Looking at the screen that came out a few minutes later, I was casual at first. "Oh, it can reproduce the look," I thought.
The double-take came after that.
The colors of the weather icons—orange for sunny, blue for rainy—and even the color-coding for precipitation probability were exactly the same. When I pressed the "Location Search / Regional Topics / Menu" tabs, the screen actually switched. When I scrolled through the time slots, the temperature and precipitation probability numbers moved in sync.
A "moving thing" came out of what should have been a static screenshot.
That's where I stopped.
There was one more thing that gave me goosebumps. Finally, Fable 5 returned a self-evaluation. I had asked for the self-evaluation, but the content was more honest than I imagined.
It said, "Self-evaluation: 88 points. Since I can't use the actual images from ads, I simulated them with CSS." It even confessed the specific reasons for the point deduction.
It didn't just copy the appearance. It inferred the mechanism behind the screen and reconstructed it. The moment that hit me, I was brought back from the side of those who had given up.
Let's try it (just copy and paste)
It's not difficult. If you follow the steps, you can definitely do it.
[Step 1: Switch to Fable 5]
Open the Claude app or browser. Tap the current model name near the input field. Just select "Fable 5" from the list.
*If it doesn't appear, please update the app to the latest version and restart. It will be reflected sequentially.
[Step 2: Save a screenshot]
Prepare one screenshot of a screen you want to reproduce. A screen from an app opened on your smartphone is fine. This time, we used the Yahoo! Weather forecast screen.
*Please choose a screen that does not contain personal information.
[Step 3: Attach the screenshot and send the text below]
You don't need to change anything. Just copy and paste it as is.
▼ Copy and send everything below ▼
1Referring only to the attached screenshot, please create a web app that faithfully reproduces the appearance and functionality.23■ Conditions4- Combine HTML/CSS/JS into a single HTML file (in a state that works just by opening it).5- Reproduce the color scheme, fonts, margins, rounded corners, and shadows using HEX codes close to what is visible (do not replace them arbitrarily).6- Implement the buttons and menus shown so that they actually work when pressed (no dummies; maintain state with JS).7- Naturally supplement parts that are not shown to match the overall tone.89■ Report after completion10- Self-evaluation of reproducibility (out of 100 points) and a breakdown of deductions.11- Parts supplemented by inference and the reasoning behind them.1213First, briefly organize the elements you were able to read, and then generate the code.
Just wait a few minutes. Here is the result when I actually tried it. Isn't it crazy? A screen with both the look and movement reproduced will appear before your eyes. And finally, Fable 5 will return its own score. The "88 points" moment I experienced will happen on your screen too.
Please post the resulting screen by quoting this article. I will repost all of them.
The story of the "Safety Device" on this crazy model
When it's this strong, you might worry: "Won't it be misused?"
The official source has an answer for that concern. Fable 5 has a mechanism called "Automatic Fallback."
It sounds complicated, but what it does is simple:
・A dedicated check AI constantly monitors your questions.
・It detects only dangerous questions, such as cyberattacks or biochemical threats.
・Only for those questions, the model one level below, Opus 4.8, answers instead.
You might think it sounds inconvenient. But according to official data, this is triggered in less than 5% of sessions. In normal use, you will almost never encounter it. Moreover, this safety device is quite solid. They did over 1,000 hours of breakthrough testing before release, and not a single loophole was found. You can use it with peace of mind.
It can do much more
The weather forecast reproduction is just a small part. Here, an honest doubt should arise: "Maybe it just worked by chance?"
I understand. But remember, the person who made that screen was none other than you. You didn't need special settings or coding knowledge. And the same thing is happening one after another with things other than weather apps.
Right now on X, various people are posting things they made with "one prompt":
・Falling block puzzle games like Tetris or Puyo Puyo
・Mario-style side-scrolling action games
・Board games like Othello or Shogi where the opponent (AI) actually moves
・Solar system simulators that calculate planetary orbits and even predict the next solar eclipse
・Physics simulations that reproduce the swinging of a pendulum or the movement of water and cloth
・Practical tools you can use immediately, like Pomodoro timers or household account books
・Data conversion that reads graph images, digitizes them, and recreates them as tables or graphs
None of these require specialized knowledge. With one prompt to copy and paste, a moving thing comes out. It's not just "mimicking the look." It calculates, moves, and creates the whole thing to the point where you can play with it. This is the power of Fable 5.
And here is the main point:
・Hand over your own X analytics screen and have it create an operational analysis dashboard.
This is the real goal. Not someone else's example, but the moment the AI works with your own numbers. That's when it first changes to "This can be used for my job."
AI's work changes from "Tasks" to "Responsibility"
Finally, a deeper story. For you who have read this far and moved your hands, it should reach you now. Felix Rieseberg, who leads the development of Claude Code at Anthropic, has declared:
"A major shift from the era of giving AI tasks to the era of giving it responsibility is beginning."
What does this mean? Until now, if an app had a bug, he would ask the AI to "find the cause." This is a "task." You do it once and it's over.
Now it's different. He leaves the job of constantly monitoring for bugs 24 hours a day to Claude. This is "responsibility." From a one-time "request" to a "job" left entirely to the AI.
The reason Fable 5 can operate autonomously for several days is for this kind of usage.
However, there is one thing we must not forget. Even if you leave the job entirely to the AI, the final responsibility for the results remains with the human. This doesn't change, and it shouldn't be changed. If so, what will be questioned from now on is this: How much work will we leave to the AI, and what will humans continue to take responsibility for? The timing for each individual to start thinking about that line has finally come.
Frequently Asked Questions
Q: Can I use it with the free plan?
A: The official guidance is for Pro/Max/Team/Enterprise plans. There is no announcement regarding provision to the free plan at this time.
Q: Do I no longer need to use Opus 4.8?
A: No. For short questions or daily tasks, there is almost no difference. Using Fable 5 only for "long and complex jobs" is a smart way to differentiate.
Q: Why does it sometimes say "Opus 4.8 responded"?
A: That's the safety device fallback. It's not a malfunction. It triggers in less than 5% of sessions.
Q: Is the conversation content used for training?
A: In the Mythos class, data is retained for 30 days for safety monitoring, but the official statement clearly says, "We do not use it for model training."
Q: What happens after July 8th?
A: Subscriptions will require usage credits. The official policy is to return it to the plan standard once capacity is ready.
Summary
Good job. Let's look back at what we learned today:
・Fable 5 is the first publicly available "Mythos-class" model above Opus.
・80.3% on SWE-bench Pro. A huge lead over other models.
・The true value is "leaving long and complex jobs entirely to it." Stripe completed two months of work in one day.
・Safety design that automatically switches for dangerous topics.
・No additional cost for Pro/Max/Team until July 7th.
For short questions, there isn't much difference from previous AIs. The difference appears when you hand over a heavy job you were hesitant to do because it was tedious in its entirety.
If you do just one thing today, let it be this: Attach just one screenshot, weather or anything else, and send the copy-paste from the article once. Don't just end with reading and understanding. Go as far as touching it and doing a double-take. I want that moment when I did a double-take at the screen to happen on your own screen.
Beyond the "Double-Take"
To those who were surprised by the weather forecast: I don't want you to let that one time end as a "fluke." I want you to proceed to the conviction that "This is something that can do anything."
For that purpose, I have prepared something special: "AI Capture: 32 Deluxe Bonuses." And they are all free. The way to receive them is simple. It starts with joining the LINE Open Chat below.
Open Chat "Fuji AI [Latest Information Distribution]"
You can enter from here.
Let's go all out and touch it until July 7th. I will continue to write very easy-to-understand articles about the latest AI. If you found this even slightly helpful, please follow me. If you give me your impressions in a quote, I will repost all of them. You, who once gave up on AI, which will you become in this one week?
