20:35AI Edge · Tutorial
How To Actually Use Claude Fable (Without Hitting Your Limit)
A 20-minute breakdown of the 10/80/10 system and loop engineering — the cost-efficient way to run the most expensive AI model on the market.
June 11thThe argument in one line.
Before an expensive AI model is taken away, you can extract its reasoning style into a written handover document and transplant that thinking into a cheaper model, preserving most of the value at a fraction of the ongoing cost.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude regularly for business or coding work and want to keep getting high-level strategic output after losing access to a premium model tier.
- You run repeatable AI workflows (skills, prompts, loops) and want to upgrade their quality with minimal ongoing cost.
- You're deciding whether to pay for AI usage credits or an API key and want a concrete sense of what different usage patterns actually cost.
SKIP IF…
- You don't use AI models with tiered/premium access at all.
- You're looking for coding tutorials rather than a workflow/cost-optimization strategy.
The full version, fast.
A premium AI model is being pulled from standard subscription access, so before that happens, the host has it write a detailed handover document describing exactly how it thinks — how it interprets requests, breaks down problems, verifies its work, and communicates conclusions — framed as a retiring analyst's brain dump to a successor. That document is then handed to a cheaper model, which noticeably improves its output quality without needing the premium model itself. For ongoing use, the recommended pattern is a 10-80-10 split: the premium model only handles the opening planning and closing review of a task, while a cheaper model executes the bulk of the work, cutting daily cost from $20-40 down to roughly $3-10. An "advisor/executor" architecture automates this by having the cheaper model call the premium one on demand mid-task rather than running everything through it.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:35
01 · Intro & the cutoff
Fable is coming off subscriptions July 7 (US) / July 8 (EU); Anthropic says it's too compute-intensive to offer standard, moving to a cost-per-credit model.
01:35 – 02:49
02 · Managing usage credits
How to find usage credits in account settings and the risk of auto-topup silently burning tokens if not monitored.
02:49 – 07:42
03 · The extraction loop
Core method: Fable writes a handover document of its reasoning style, which is fed into Opus 4.8 or Sonnet 5 to upgrade their output on recurring skills/workflows.
07:42 – 09:38
04 · Demo: updating a skill
Live demo feeding the handover document back into Opus to update an existing 'video production pipeline' skill.
09:38 – 11:02
05 · Handover into project instructions
Alternative: add the handover document to a project's persistent instructions so every chat in that project benefits.
11:02 – 12:34
06 · Building a new skill via interview
Second technique — ask the model to interview you one question at a time about a workflow, then output a complete skill; run /skill-creator.
12:34 – 14:22
07 · The real sauce
Framing Fable as an adviser/orchestrator rather than an executor — use cheaper models to do the actual work.
09:22 – 11:23
08 · What it costs
Cost breakdown: extraction ~$3 one-time; big tasks $10-20/session; full daily-driver use $20-40/day at $10/1M input, $50/1M output token pricing.
11:23 – 13:58
09 · The 10-80-10 rule
Premium model handles first 10% (plan) and last 10% (review); cheaper model executes the middle 80%, cutting daily cost from $20-40 to $3-10.
13:58 – 14:22
10 · Outro
Recap and teaser for a follow-up video on using loops.
Lines worth screenshotting.
- A one-time ~$3 prompt can extract a premium AI model's reasoning style into a document that measurably improves a cheaper model's output.
- The extraction prompt works by asking the model to write a handover document as a 'retiring senior analyst's brain dump to their successor' — a way of thinking, not a list of rules.
- Running every task through the most expensive model in a daily workflow can cost $20-40/day, versus $3-10/day using a 10-80-10 split.
- The 10-80-10 rule: the premium model handles the first 10% (planning/orchestration) and the last 10% (review), a cheaper model executes the middle 80%.
- An 'advisor/executor' architecture lets a cheap execution model call an expensive model on demand as a tool, rather than running the whole task through the expensive model.
- A handover document extracted from one model can also be fed into a project's persistent instructions, so every conversation in that project inherits the upgraded reasoning.
- The same interview-style prompting pattern used for extraction can also build a new reusable skill: have the model ask one question at a time until it fully understands a workflow, then write it up as a complete skill document.
- Bigger coding sessions on a premium-tier model can run $10-20 per session, and a full day of heavy use can reach $50-100+, which is why selective use rather than default use matters.
- Downgrading a subscription tier and paying the difference in pay-per-use credits for only the highest-value calls can be cheaper overall than staying on the top subscription tier.
Extract expertise before you lose access to it
WHAT TO LEARN
When a premium tool or resource is about to become unavailable or unaffordable, you can often capture its most valuable output — its way of thinking or working — in a portable document rather than losing the value entirely.
- Before losing access to any premium resource (a tool, an expert, a subscription tier), capture its distinctive value in a transferable document rather than assuming the value disappears with the resource.
- A well-designed one-time extraction (a detailed writeup of 'how' something works, not just 'what' it concluded) can be worth far more than repeatedly paying for the resource itself.
- Framing a request as 'explain your reasoning to a successor' produces a fundamentally different, more useful document than asking for a simple summary or a list of rules.
- Reserve your most expensive resource (time, money, expert access) for the highest-leverage parts of a task — planning and final review — and let cheaper resources handle routine execution.
- A 10-80-10 style split (expert plans, junior executes, expert reviews) can cut costs 70-90% versus using the expensive resource for everything, with minimal quality loss.
- Build reusable playbooks by interviewing yourself or an expert one question at a time until the reasoning is fully captured, then write it down as a repeatable procedure others can follow.
Terms worth knowing.
- Extraction loop
- A one-time prompting technique where a more capable AI model writes a handover document describing its own reasoning approach, which is then given to a less capable model to improve its output.
- 10-80-10 rule
- A cost-optimization pattern where an expensive AI model only handles the first 10% (planning) and last 10% (review) of a task, while a cheaper model executes the middle 80%.
- Advisor/executor architecture
- A workflow where a cheaper 'executor' model runs the main task loop and calls a more expensive 'advisor' model on demand for strategic input, instead of routing the whole task through the expensive model.
- Usage credits
- Pay-as-you-go credits some AI subscriptions let you buy to keep using a model after hitting your plan's included usage limit.
Things they pointed at.
02:45productClaude usage credits (account settings)
02:23linkplatform.claude.com/dashboard (API portal)
07:40toolWhisperFlow (voice transcription tool)
08:20tool/skill-creator (Claude Code slash command)
11:43toolArtificial Analysis (model benchmark site)
Lines you could clip.
00:04
“Anthropic, the creators of Claude, are about to pull Claude Fable, the most powerful model in the world, out of your subscription.”
05:05
“You are the most capable model I have access to. I'm about to lose access to you. But before that happens, I want you to write a complete handover document for your replacement.”
12:00
“I've always said that you should use other models and use Fable as an adviser slash orchestrator and then use Sonnet or Opus as an executor anyway.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
00:00Anthropic, the creators of Claude, are about to pull Claude Fable, the most powerful model in the world, out of your subscription. I spent today , which I can use to turn Opus 4.8 into a Fable like model.
00:16So if you're watching this before Fable has been taken away from you, you're gonna be able to do the exact same. If you're watching this video after Fable's being taken away, don't worry because I'll show you exactly how to do the same thing for just a few dollars. So today, I'm gonna give you the exact prompt.
00:31I'm gonna show you exactly how to get access to Chord Fable again once it's taken away in your settings. And then I'm gonna show you the best way that you can use Chord Fable after it's out of your subscription for the most cost effective manner if you're a power user.
00:47So if you wanna know when Claude Fable is going out of subscriptions, it's essentially gonna happen on July 7, but this will likely be in Europe early on July 8. We're not sure exactly how long it's gonna be out of subscriptions.
00:59My best guess would probably be a few weeks, but one of the official developers at Anthropic said that while it's coming off subscriptions after July 7, they're going to aim to restore Fable as a standard part of their subscriptions as soon as capacity allows as we mentioned in our original blog posts. Essentially, issue Anthropic is having right now is Fable is so expensive to run.
01:20It's so compute intensive that they just can't afford to offer it. It's just not feasible to all their subscribers. That's why it's switching to a cost per credit model.
01:29So if you wanna keep using it, you will need to pay for credits. But of course, today, I'm gonna show you the hack to alleviate that for a lot of you. So what you wanna do after it's taken away is you want to go into your settings in the Cloud app or the Cloud website, click on usage, and then go down to usage credits.
01:45If you wanna automatically continue to top up credits as you use Fable, you can switch this on. If you switch on this toggle, it'll keep topping up your account with usage credits after you use your plan limit.
01:55But you have to be careful here and keep actively checking your usage limits in your settings, or you could burn a lot of tokens with Fable. That's why also later in today's video, I'm gonna show you the best way to use Fable, and this is actually a secret that a lot of the pros are using to get the most out of Fable without breaking the bank.
02:11And it's the exact system that I'm currently implementing across my business, which is a 30 person media company where we produce thousands of videos and run entire developer operations. So we need to use this or we'd burn probably like $50,000 a month on Claude Fable alone.
02:24And if you're a serious power user of Claude, you can go to platform.claude.com forward slash dashboard, and you can essentially go to the API portal and get an API key, which will enable you to access Claude Fable this way.
02:37Also, later on, I'm gonna talk about how much it would actually cost to run Claude Fable this way, either through the usage credits or through the API, because I'm seeing a lot of misconceptions there. But firstly, let's go through right now how you can extract the power of Claude Fable using one prompt.
02:52And I'm using this exact extraction loop on every single one of my skills. So all the repetitive processes, my loops, my skills that I use to complete business activities. For example, business development scraping.
03:04For example, video production. So script writing, short form content writing, x writing, and my business strategy skill in general, which I use inside my business memory system. So if you're watching this video early, you're in luck because you can do this for free right now as a part of your subscription.
03:18If you're watching after the cutoff, this is probably only going to cost you a couple of dollars worth of credits, and then you're gonna get a lot of the power of Fable for potentially the month or the few weeks that we don't have access to Fable within Opus 4.8 or even the new Sonnet model. So this is essentially the extraction loop.
03:35It starts with Claude Fable five. Obviously, you'll need access to the model for the case of this prompt to actually do this. Then you wanna create a handover document.
03:42It might cost a few dollars if you're watching this after. If you're watching this before, great. It's pretty much gonna be free as part of your subscription.
03:47You wanna give this document to Claude, which I'll show you, and you wanna run it through Opus 4.8 or Sonnet five, whatever model you want to use, we'll go through that as well. And then you wanna apply that to your workflows. So your workflows, your skills, your repetitive processes will all get the benefit from Fable like intelligence without having to run the same prompt over and over and over again on Claude Fable, which could burn tokens.
04:07I mean, this as a daily drive, it could cost you between 20 to, let's say, $40 a day if you were doing that every time. Okay. Let me show you an example.
04:15So I have a skill called video production pipeline. This is the skill that I use every time I film a video And after a video comes out based on my team's feedback and my own analysis and the data, I give it a feedback loop. So I tell Claude what to improve.
04:28So every single time it gets better at planning videos for me. A video like this, I don't really lean on it too much because it's mostly just off the top of my head, but the really complex videos, the trading bot stuff, etcetera, obviously, it helps to have Fable as a thinking partner and as an assistant. So let's assume I'm about to lose Fable, and I'm gonna have to use Opus 4.8 to execute this skill in future.
04:48What I wanna do is I wanna use Fable and I wanna enter this prompt. You are the most capable model I have access to. I'm about to lose access to you.
04:56But before that happens, I want you to write a complete handover document for your replacement, a less capable model, aka Opus or Sonnet, that will take over your job. Write it as a retiring senior analyst brain dump to their successor. Not a list of rules, a way of thinking.
05:10So it's gonna capture the way Fable thinks. How to interpret what a request is really asking for, how to decompose problems, how to verify work instead of pattern matching, how to communicate conclusions, and how to self review before answering. Be exhaustive.
05:23Every section must earn its place. Enter. And by the way, if you click the link in the description down below for free, I'm gonna give you access to the entire guide from today's video that you can actually just drag and drop straight into your cord so you can do this really, really easily.
05:36Just follow the instructions on the pin link down below. Literally takes ten seconds and you're gonna get access to everything from today's video. And by the way, going forward, that's what I'm doing for every single video.
05:45So if you ever want the assets or the extended information from video, it's always accessible down below. Okay. So what it's done is it's created a handover document.
05:54So for this skill that I routinely run on Fable, it's created an entire brain. So what I would then do is I would download this file, and then what you wanna do is you wanna go into Opus. Let's say once you lose access, and you wanna say something like this.
06:10I'll use WhisperFlow, the a voice transcription tool, by way. It's amazing. You should definitely get it.
06:14Not sponsored, which I was. It's seriously great. So watch this.
06:18Hey. Could you please analyze this file, and could you please update the video production skill with the new information from this document to make it even better?
06:30And then, of course, you'd take the skill. And skills are also a way for you to train loops. So even if you run a looping strategy, they're often still trained on skills.
06:39The other thing you can do is if you have a project folder, you can actually add a set of instructions. So this wouldn't be for a specific skill. This would be if you use the prompt for, let's say, a strategic output for a business you run.
06:52You could add the file to the instructions. So every time you message a chat within that project, you're getting fable like intelligence. Now, of course, I'm gonna caveat this myself.
07:01This is not a system to actually get the exact Fable like intelligence on every single prompt. What you're doing is you're extracting one time the way Fable thinks in order to make Opus think better. So it'll definitely improve Opus.
07:14And in the tests I've run, the outputs are definitely better. They're definitely more strategic. But obviously, a lot of the nuance that makes Fable so good is like those little extras that it will give you, let's say, a brainstorming session or, you know, when executing on code.
07:28You know, you can't do a document and put it into Claude code and say, hey, code like Fable, because Opus literally isn't capable of that. But this is something that you can do to make your normal workflows better on Opus. What I'm gonna get to very shortly in today's video, which I think is honestly the real source of today's video, is the actual strategy you can use to use Fable when it counts, but saving on, you know, the extreme cost that you would pay by using it all the time.
07:53So those willing to pay a little bit of money on credits are gonna get immense benefit. I'll walk you through this, and there's a tool that I use which is , which I'll also give to you today. So you can see here it's actually editing the skill.
08:04I'll give you another example as well. You can actually create a skill. So for example, a prompt like this, interview me about a certain workflow.
08:11So interview me about x copywriting. Let's say I'm creating a a skill to help me write one question at a time until you fully understand how to do it, then write it as a complete skill.
08:24So the AI assistant could follow the skill, and we're gonna do this on Fable while we have access to it, or later you could pay a few dollars for this, and probably less if it's just one prompt, and you could go forward slash skill creator. So it will actually create a new skill. This is how you build skills, by the way.
08:37You click enter. And by the way, what I recommend is having a local folder. You can also have a cloud system, but for a simple way, have a local folder with a library of all your skills.
08:46I mean, obviously, Claude saves a library anyway, but I like having it all in one place so I can just, you know, use skills all the time whenever I'm prompting something regarding my business strategically. So here it's gonna ask me questions. You can use Whisper Flow or voice commanding, so it actually transcribes every single thing you're saying, so you can answer the question.
09:01I used a hotkey just then, see how great that is. Such a cool tool. And I would go through that system.
09:05I would do it on Fable, and then I would take the skill created on Fable, and then I would, you know, essentially use it on Opus. That's another way that you can go about it. The link in description below will have all of the methodology for all of this, by the way.
09:17So now you know how to create simple handovers using Fable as the strategic brain. Let's go through the cost, and then let's go through the best way that you can actually use Fable after the cutoff. So the cost of Fable is obviously immense.
09:31We know it's one of the most expensive models in existence. It is the smartest model as you can see, one out of one seventy, but it is the thirteenth most expensive model as well, which, you know, out of a 170 models, that's, you know, extremely extremely expensive.
09:43Essentially, it's gonna cost you for, like, a simple extraction probably a few dollars. So if you did that process across a few skills using the credits that we discussed before, it's probably gonna cost you a few dollars. But if you were to use really big tasks, so let's say, like, you were coding something, it might cost 10 to $20 per session based on the math of the cost per token as you could see here, which is $10 per million.
10:04A million's a fair chunk, but you will burn through it if you're actually, you know, coding something like a game or a dashboard or a website or something. And that's just for an input for the actual outputs. It's $50.
10:14So it does rack up a fair bit of cost. Now, obviously, creating skill or basic workflows or basic loops isn't gonna cost that much. I don't recommend that you run execution on Fable anyway.
10:24I've always said that you should use other models and use Fable as an adviser slash orchestrator and then use Sonnet or Opus as an executor anyway. So the way you would actually use Fable in practice beyond just the extraction method I talked to you about is using it as the strategist ten eighty ten rule. So first 10 percent's Fable, 80%.
10:43So if you're coding that, for example, Fable orchestrates the plans with the first 10%. The middle 80% is going to be the execution layer that's done by Sonnet or Opus. Honestly, Sonnet is very capable now, and it's cheaper.
10:54The the the new version's absolutely great. And then you'd use Fable for the last 10%. The review, the tweaks, etcetera, that's gonna get your cost if you do on a daily drive Fable down from $40 to probably a few dollars, maybe 3 to $10 a day.
11:06If you're serious about AI, it's probably worth it. Like, it's better just not to drink coffee or maybe coffee gives you a huge productivity boost when you use Claude coffee plus Claude Claude is probably the best combo. But just for the price of a coffee, you get the benefit as Fable as a strategic adviser.
11:19So maybe it's even worth actually lowering your subscription overall, which is like, you know, if you're on the max plan, it's like $200, lower your subscription down to the $100 subscription and then, you know, use Sonnet, a cheaper model, and then use Fable and pay the extra $100 in credits throughout the month and get 3 to $5 a day.
11:35That could be a strategy you use to cost optimize. If you don't mind spending extra, it might just be worth it to suck it up for a couple of weeks. Obviously, completely dependent on your situation and your budget until until we get Fable back again, which we know at some point is gonna happen.
11:47But this is the hack I wanna show you, and this will also be available in the document below. I'll send you the link to this, and it'll be very easy. You can also look it up yourself.
11:54It's platform.core.com/docs/en. I mean, you can just look at the the URL in the video, of course, forward slash agents and tools hyphenated forward slash tool hyphen use forward slash adviser hyphen tool. This is the adviser tool.
12:08Obviously, I'll leave it below so you can copy the page. And this is essentially a prompt that you can run to use Fable as an orchestrator or an adviser. So here's how it works.
12:18Fable is your adviser. It's your on demand strategic analyst going back to what I said before. This is basically a practical automatic way to implement the 10/8010 strategy that I've It will basically act as an on demand adviser.
12:30It'll send advice to your context system, so either the inbuilt Claude memory or, course, if you have a folder system, it'll update the MD files there with your conversation tools history, etcetera. Then it will send that data to Sonnet. Sonnet will execute a loop if you use forward slash loop, which is now the best way to use Claude.
12:44I'm gonna do a video on it. It will essentially run the strategy for you, and then it will call Fable whenever it needs advice. So at the beginning or at the end of a process, and you can specify how you would like that to happen.
12:56This is the best way to run Claude right now, the most cost effective way to run Claude right now and after the cutoff. It's absolute gold, this strategy. And I'll, of course, also leave it in the document below.
13:06But the adviser tool is amazing because essentially what you can do is you just copy it in on Python using quick start. You can even give Claude code the link to this document, and it will just install it itself. And essentially, this is how it works.
13:17An executor emits a server tool use block with the name adviser and an empty input. The executor signals timing and the server supplies context.
13:26So this will run automatically on your behalf. I'm not gonna bog down this video and make it too developer heavy because Claude can actually distill a lot of this information for you and you can ask it questions, but this is essentially how it works. And this is how a lot of the pros are using Fable right now.
13:39They're getting the bang for their buck out of the intelligence, but they're not wasting tokens on executing. Because Fable is actually quite slow at executing. I mean, look at the speed here.
13:48It's ranked 72 out of a 170 models. Sonnet's faster. Opus is faster.
13:52Use those models to execute fast, and then use the intelligence where Fable's ranked one out of one seventy where it really shines for anything strategic, and that is is gonna be how to get the most out of the model.
14:04And that's exactly the strategy I'll be executing. I'm gonna do a video on loops because I feel like that is the next step to build upon, but I wanted to get this video out before Fable has taken off your subscription so you know exactly what to do. The link to the document will be available below, and I will see you in the next video.
14:19Subscribe for more content like this. Have a lovely rest of your day. Peace out.
The Hook
The bait, then the rug-pull.
A premium AI model is about to disappear from standard subscriptions — so the host spent a day extracting its "brain" into a document he can hand to cheaper models forever. What follows is both a preservation trick and a broader lesson in not paying premium prices for work a cheaper tool can do just as well.
Named ideas worth stealing.
03:29model
The Extraction Loop
- Claude Fable 5
- Handover Document
- Opus 4.8 / Sonnet 5
- Your Skills
Have the premium model write a one-time handover document of its reasoning approach; feed that document into a cheaper model's skill or project instructions to durably upgrade its output quality.
Steal forany recurring AI workflow/skill you want to upgrade without paying premium-model prices every run
10:23concept
The 10-80-10 Rule
- First 10% — premium model plans
- Middle 80% — cheaper model executes
- Last 10% — premium model reviews
Reserve the expensive, high-intelligence model for orchestration and final review; let a faster, cheaper model do the bulk execution.
Steal forany multi-step AI task where cost scales with model choice, e.g. coding sessions, content production pipelines
12:43model
Advisor/Executor architecture
- Executor (cheap model) runs main loop
- Executor calls Advisor (premium model) via tool call on demand
- Both read/write a shared common context
Automates the 10-80-10 rule: instead of manually switching models, the cheap executor model calls the premium advisor model as a tool whenever it needs strategic input, keeping both in sync via shared context.
Steal forbuilding agentic workflows that need occasional expert input without constant premium-model cost
How they asked for the click.
VERBAL ASK
05:25newsletter
“click the link in the description down below for free... you can actually just drag and drop straight into your cord... follow the instructions on the pin link down below”
soft mid-roll CTA repeated near the outro; gates the actual extraction prompt/guide PDF behind a free newsletter + WhatsApp community signup — a lead-gen funnel disclosed in description.txt
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
