CODE IN TOKYO

Hands-on Review: Claude's New Fable Model Is Much Better Than I Expected

I Tried Fable in Real Work

I put Claude’s new model, Fable, into actual work.

I had a task to build a new feature, so I used it experimentally.

My short conclusion: I was genuinely surprised.

I have only been using it for a short time, so my view may change later. Still, I want to write down my current impression.

It Can Go from Design to Documentation in One Flow

This was the first thing that impressed me.

Until now, I usually broke work into separate stages:

With Fable, however, I can hand over a reasonably large block of work, and it can turn that into something concrete.

From a simple explanation like “I want to build something like this,” it can take the task surprisingly close to completion.

My impression is that Fable works better when you give it a whole piece of work rather than feeding it many tiny tasks one by one.

It Misunderstands Less Often

This was the part that surprised me the most.

When using AI, there are moments when I think:

“No, that is not what I meant…”

This happens between humans too, but it happens even more often with AI.

Fable, however, has far fewer of those mismatches.

Of course, it is not perfect.

But in my experience so far, its ability to understand the intent behind a request is very strong.

To be honest, there were several moments when I thought:

“Claude Opus might have misunderstood this.”

I did not run an intelligence test, so I cannot make a definitive claim. But based on how it felt in real work, Fable is extremely capable.

It Is Not Ideal for Small Tasks

On the other hand, its strength is harder to see with small jobs.

If I only want to change one line.

If I only want to adjust a small configuration file.

If I only want to know the cause of an error.

For those uses, Codex or Opus is probably enough.

Fable starts to show its real value when the request is more abstract:

“Design this.”

“Create a mockup too.”

“Think through the whole structure.”

In other words, using Fable feels close to handing a large task to a highly capable engineer.

The Only Problem Is Cost

I have praised it a lot so far, but there is one problem.

The price.

Fable is not unlimited. It uses a credit-based system.

That matters more than I expected.

A small investigation.

A light consultation.

A quick check that is almost like casual conversation.

If I use Fable for those things, the credits keep disappearing.

And then, strangely enough, I start asking myself:

“Does this really need to be done with Fable?”

I know the performance is high.

But the wallet matters too.

My Current Conclusion

So my future usage will probably look like this:

In short:

“Use it only when it really matters.”

Fable is not an everyday hammer.

It is more like an Elixir in an RPG.

It is reassuring to have.

If you use it, every status gets restored.

But it is expensive, so you keep saving it. I feel like I might finish the game while still holding onto it.

Well, in about six months, it will probably become usable without extra billing. So paying for it only when necessary may be the right approach.