CODE IN TOKYO

Fable 5 Is Still Smart. How Much Easier Can a Considerate AI Make Human Work?

Fable 5 Is Available Again for Claude Code Users

Fable 5 has become available again for Claude Code users, apparently for a limited time. At the moment, it seems to be usable until July 7, within the weekly usage limit.

So I immediately gave it a fairly heavy refactoring task.

Here is my conclusion.

It is still more considerate than Opus 4.8.

This is not about benchmark scores.

The difference is in how easy it feels to work with the model after spending hours developing together.

With Opus 4.8, there are still many situations where the human side needs to lay down detailed rails:

Of course, Opus 4.8 can still do very good work when given that context.

But Fable 5 seems to infer much more of that surrounding intent.

Even when I explain my goal roughly, it often responds as if to say:

“You probably want something like this, right?”

And then it brings the work surprisingly close to what I expected.

The Word Is Considerate

If I had to describe this difference in one phrase, it would be:

“It is considerate.”

As a result, the human side becomes much more comfortable.

I need to write fewer prompts, and there are fewer points I need to review in detail.

The management cost of getting AI to do useful work definitely feels lower.

But It Uses Tokens Quickly

After using it for a full day, I also noticed something else.

Because it is considerate, it also uses tokens generously.

It thinks across a wider range, so the allocated tokens decrease quickly.

To be fair, I was also giving it heavy tasks because I wanted to take advantage of the opportunity. In normal daily use, the situation may not be quite as extreme.

Still, if you think, “This is easy, so I will let Fable 5 handle everything,” the usage limit can start to feel uncomfortable very quickly.

Of course, if a human writes careful instructions, Opus 4.8 can usually reach the same quality for ordinary programming tasks.

The difference is not simply capability.

It is the amount of human effort required to reach that quality.

Do I pay money to reduce that effort?

Or do I work a little harder myself?

That is ultimately the difficult decision.

My conclusion after using it for a day is this:

“It is definitely smart. But for now, it is probably better for my wallet if I work a little harder myself.”

For a while longer, it seems the human side still needs to make an effort.

Contact

I provide system development and technical consulting using AI, AWS, and Claude Code.

Contact form: https://holly-money-e94.notion.site/390ff30cf18c8086a676fe630d171873