Fable 5: FREE Tools Cut Tokens by 95%!

**Julian Goldie** (0:00)
Today, I'm going to show you some of the best ways to reduce the amount of tokens you actually use inside Fable 5, because Fable 5 is absolutely awesome. But if you're not careful, you're going to absolutely rinse through the tokens you have available on your subscription. Bear in mind that it uses more tokens, and you can only use 50% of your subscription on it. And also by July the 7th, everything you do doesn't go through the subscription. So now is the time to start thinking about how to reduce your token usage. The first time this came out, I actually burned through the full cap in like hours. And what we've got is a bunch of different ways and methods that you can use to quickly reduce the amount of tokens that you use. And there's 10 different moves that you can use to do this. So first of all, we're going to categorize this by route. So when you are using Fable 5 and you're building with it, this alone can cut your tokens by 50%.
Now some of the strategies I'll show you actually claim to reduce tokens by 95%. So we'll come on to those later. And this only takes a few minutes to set up. So one of the first things that you could do is stop paying the architect to lay the bricks. And what I mean by that, for example, is like every task, and every time you ask Claude to do something, you can route it to different people or different models, depending on what you want. So for example, if it's like a really hard task, like a plan, a judge, difficult bugs, then you can use Fable 5, the architect. But actually for most of the general sort of automations, Opus 4.8 is pretty good and you can just use that as a builder, the right to the actual code. And then you save your tokens because you assign your tasks to Opus for the hard stuff, Fable 5 for the really, really difficult stuff. And then Haiku even can do like the small sub-agent stuff, or you can use something even cheaper like Glm 5.2.
So for example, we've got Glm code here, which is basically Claude code running Glm 5.2. So that if we have, for example, some harder tasks, some more work, etc., we can run it through there instead. Now, additionally, move number two is that Fable can plan and Opus actually builds for you. So this way you get Fable 5 to build the blueprint, but not for the actual type, not for the actual coding. Because Opus uses less tokens, you get more tokens available with it. So how does that work? Well, basically you would get Fable 5 to draw the blueprint with plan mode. Then you get the plan, and this is like all the decisions forked, and then Opus actually lays the bricks for you. When I say lay the bricks, I mean like actually go off and do the work for you. So we actually did a playbook on this already, but basically if you plan with Claude Fable 5, you implement with a cheaper model, based on Kilo Code's research, you can produce the same service for 59% less than if you're using Claude Fable 5 for both phases. So the way this works is you have the planner, the genius, which is Fable 5, then your agents can build with smaller, cheaper models. So that alone can reduce your cost by 50% as well. Now you can also change the level of effort. So you can change the effort dial based on what you need to do. Now you've got a bunch of choices here. So if you go into Claude, you can change what it runs on. So it can go from like low to medium to high to extra, max and ultra code, right?
So you don't really need to go too high on this, and it depends on the task. Like if it's just something simple like a blog post, you can go pretty low with that, right? But if it's something super difficult, like for example, maybe you are coding out a really difficult bug, or you're trying to fix something in your app, then you'd go over here. But the main point here is like, you can reduce the amount of effort you actually use based on the job, which means you use less tokens when you're running through Fable 5
Next up, there's a really good tool called Headroom. This is free and open source.
And the way this works is that this shrinks what Claude code sends automatically in every session. So this is a tool you can ask Claude just to install it for you. It will reduce the amount of tokens you use. Now on their own tests, and just try this yourself, I've tested it out myself as well. I got another tutorial on this. But basically Headroom can reduce your tokens by 60 to 95 percent. So you can save your tokens by 60 to 95 percent, which is pretty wild. So you imagine you're combining all of these different methods.