5 Fable 5 Use Cases to Run Before It's API-Only
5 Fable 5 Use Cases to Run Before It's API-Only
If you've got cheap access to Claude Fable 5 right now, spend it on these five things before the window closes. Once you're paying thousands on the Fable 5 API instead of running it inside your plan, most of these stop being casual experiments and start being budget decisions. So here's exactly what I'd knock out in the last 24 hours — planning, a second brain, two audits, and one all-out token burn — with the prompts I actually used.
Before the use cases, one thing about usage. Don't sprint through the first two and blow your whole allowance. Pace it, because use case five is where you go crazy on purpose.
What Effort Level Should You Run Fable 5 At?
Lower than you think. On the DeepSuite benchmark, Fable 5 on medium — or even low — beats Opus 4.8, and it does it for less money.
The cost gap is real. On medium, Fable 5 comes in around half the price of Opus 4.8 on the same benchmark run (roughly $6 vs $13). Drop to low and the gap widens further (roughly $3.76 vs $13). You do not need max or extra-high effort to get frontier-level output from this model.
So the play is simple: run the first few use cases on medium, dip to low if you're tight on usage, and only ramp the effort up near the deadline when you're trying to burn everything you've got left. That way you get maximum output per token instead of torching your allowance on effort levels you didn't need.
How Do You Use Fable 5 to Plan Work for Opus 4.8?
This is the most obvious move and the highest floor: use Fable 5 as the planner and let Opus 4.8 be the executor down the line.
Point Fable 5 at a codebase or project you're already working on. It reads the thing top to bottom, figures out what needs to change, and writes out what Opus 4.8 should actually execute on the ground to reach a sensible end state. The prompt I use has it rank the top five changes by leverage and tell me what to do first — but you can uncap that and have it surface every change worth making, with the reasoning for each.
When I ran this on one of my codebases, it came back with a specific plan per major issue — five separate markdown files, each one a hand-off spec I can drop straight into an Opus session. That's the pattern: Fable 5 thinks, Opus 4.8 builds, and the plan lives in version-controlled markdown so nothing gets lost between the two.
It works on greenfield too. Edit the prompt slightly and it'll scope out a brand-new project the same way instead of auditing an existing one.
How Do You Build a Second Brain With Fable 5 and Obsidian?
Use case two: have Fable 5 stand up a second brain in Obsidian, based on Andrej Karpathy's post on LLM-friendly knowledge bases.
Here's the part people miss. The pretty graph view is not where the value is. The value is giving Claude Code a map of interconnected documents so it can answer questions quickly and correctly across a huge corpus. You could have millions of documents in the vault and Claude Code still navigates it fine, because it's moving through a structured web of links, not grepping a flat pile of files.
When you point Fable 5 at this, it designates a folder as your Obsidian vault and organizes your markdown into three tiers:
- Unstructured — raw capture, dumps, transcripts.
- Structured — distilled, organized knowledge.
- Output — the shipped stuff: slide decks, write-ups, docs.
The result is a system that lets Claude Code actually know you and your projects. If you work in markdown-heavy spaces, this pays off every single day. And if you want to go further, you can wire those second-brain metrics into a full command center on screen with a terminal built in — but you don't need the OS layer to get the benefit. The second brain stands on its own.
How Do You Run a Self-Audit on Your Claude Code Sessions?
Use case three is a self-audit of your last 30 days of sessions — which, for some of you, is hundreds of sessions. You have Fable 5 mine all of it and pull out three things:
- The top five repeated patterns, with counts. What are you doing over and over?
- Three skills built from those patterns — and then it actually creates them.
- The biggest inefficiency in how you prompt. Where are you leaving performance on the table as an operator?
This is how you turn raw usage history into codified skills. The stuff you do constantly should be a skill, not a paragraph you re-type every session. The audit finds the repetition, and the same run converts it into reusable skills.
One addition I'd make that I didn't: have it compare your pre-Fable and post-Fable sessions. You might be prompting differently now, and the model might be responding differently. That said — don't fall for the "distill Fable into a prompt for Opus" trick. The model is the model. A skill that says "here's how Fable works, make Opus do it that way" doesn't actually transfer capability. The most you'll pull out of a cross-model session audit is marginal stuff about how you talk to it, not a way to fake one model with another.
How Do You Audit and Improve Your Existing Skills?
Use case four is the flip side of three. Where the self-audit asks what skills are missing, this one asks how good are the skills I already have — because skills are probably the single highest-leverage thing in Claude Code. They're how you make a non-deterministic model behave deterministically, whether that's a full workflow or just "you're bad at front-end design, here are instructions to follow."
The prompt builds a full table of your existing skills and tags each one:
- Keep — it's good, leave it.
- Merge — it overlaps with another, combine them.
- Rewrite — the logic is fine but the execution is weak.
- Delete — it's dead weight.
Then it goes after the worst offenders directly: rewrites the three worst descriptions for trigger accuracy, and fixes the worst SKILL.md body. Trigger accuracy matters more than people realize — a skill that never fires because its description is vague is worth nothing, no matter how good the body is. This run identifies the troublemakers and repairs them in the same pass.
What's the Best Way to Burn the Rest of Your Fable 5 Usage?
Use case five is the one you save for last, because this is where you burn tokens on purpose. Point Fable 5 at your wildest project using Ultra Code and dynamic workflows, and let it run until the usage is gone.
The beautiful part of Ultra Code is that you don't even need a great prompt. Start with something like deep research — have it fan out, spin up something like a hundred sub-agents, run adversarial review across them, and come back with a report answering whatever question you threw at it. Then don't stop at the plan — have it execute.
This is the opposite of the token-saving advice from every other video (Fable plans, Opus executes, advisor mode, etc.). Not today. Today you have Fable 5 do everything: research, plan, and build, inside Ultra Code, with a custom dynamic workflow it designs for the specific problem. Then you let it go to work until it's chewed through every token you have left — because it might be a while before you see access like this again.
That's the full run: plan with Fable, build a second brain, audit what's missing, fix what exists, then go all-in on Ultra Code at the end. Don't leave usage on the table.
Frequently Asked Questions
What effort level should I use with Fable 5?
Start on medium — on the DeepSuite benchmark, Fable 5 on medium (and even low) beats Opus 4.8 while costing roughly half as much. You don't need max or extra-high effort to get frontier output. Save the high effort levels for the final all-out run when you're trying to spend down remaining usage.
Can Fable 5 write plans for Opus 4.8 to execute?
Yes — that's one of its strongest use cases. Point Fable 5 at a codebase and it'll produce ranked, leverage-ordered markdown plans that Opus 4.8 can execute directly. In one run it generated five separate markdown hand-off files for a single codebase, each scoped to a specific issue.
Is the Obsidian graph view the point of a second brain?
No. The graph view looks cool but the real value is giving Claude Code a navigable map of interconnected documents so it can answer questions across a massive corpus quickly and correctly. The organization is the payoff, not the visualization — it scales to millions of documents.
Can you distill Fable 5's behavior into a prompt for Opus?
Not really. The idea that you can hand Opus a skill describing "how Fable works" and get Fable-level results is mostly nonsense — the model is the model. You might extract minor stylistic tweaks about how you prompt from a cross-model session audit, but you can't fake one model's capability with another.
What is Ultra Code and when should I use it?
Ultra Code is the high-effort, dynamic-workflow mode where Fable 5 fans out into many sub-agents, runs adversarial review, and can research, plan, and execute an entire project end to end. Use it for your biggest, wildest project when you want to spend down usage — it's designed to burn tokens for maximum output, not to conserve them.
If you want to go deeper into building with Claude Code and Fable 5, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.
