Welcome to the MonDive

Today in MonDive, we’re looking at Claude Fable 5, Anthropic’s most powerful model built for the kind of AI work that goes far beyond simple prompts.

People are already using it to build apps, websites, workflows, and full projects that usually need a lot more planning and coding. In this edition, we’ll break down what makes Fable 5 different, how to prompt it properly, and how it performs when you put it to work on a real build.

Alright, let’s dive in.

Make Claude Your 2nd Brain That works 24/7

Claude is currently the most powerful tool of 2026. It's been launching new features every week- Skills, Connectors, Cowork, vibe coding. Yet almost no one knows how to actually use them.

Our expert mentors have condensed 800+ hours of Claude research, articles, YouTube content and real-world practice into a focused 16-hour curriculum. Join the 2-Day Claude AI Mastery Workshop: a live, end-to-end deep dive into Claude plus 10+ AI tools, LLMs and workflows.

It would be silly not to SIGN UP

The professionals who master Claude right now will design carousels in seconds, ship code without being developers, and have AI working for them while they sleep.

You will learn how to:

- master Claude's three modes: Chat, Cowork and Code.

- Set up Skills, Connectors and Plug-ins to automate your desktop, Notion and files.

- Vibe code apps and dashboards without writing code & 10+ AI tools and workflows that pair with Claude.

🧠 Saturday & Sunday

🕜 10 AM – 7 PM EST

Why Fable 5 matters now

Imagine an AI model with so much power that the U.S. government had to literally step in and lock access for the entire world.

That is the kind of launch story Claude Fable 5 came with. Access was paused after U.S. export controls were applied, then restored globally on July 1. For paid Claude users, Fable 5 can be used for up to 50% of weekly usage limits through July 7 before it moves to usage credits.

Fable 5 is being positioned as Anthropic’s most powerful general Claude model yet, built for the hardest knowledge work, coding problems, and long-running projects. It is not just an AI tool that finds an answer and stops. It can check whether the answer is good enough, improve it, and keep working through the next part of the task.

That makes it useful for bigger projects like:

• building games, websites, and app prototypes

• working through complex coding projects

• understanding charts, diagrams, PDFs, and documents

• planning multi-step workflows that need more than one quick response

This is the type of work companies are willing to pay serious money for, because the hard part is not always writing the final output.

The hard part is deciding what needs to be built, what can go wrong, how the pieces connect, and how to keep improving until the result is actually useful.

Fable 5 Against the Frontier Models

Fable 5 does not just look stronger against GPT-5.5 and Gemini 3.1 Pro. The bigger story is how much it moves ahead of Anthropic’s own Claude Opus 4.8 in the areas Fable was built for: agentic coding, long-running tasks, tool use, document reasoning, and complex knowledge work. In Anthropic’s official benchmark table, Fable 5 leads Opus 4.8 on key rows like SWE-bench Pro, FrontierCode Diamond, GDPval-AA, GDP.pdf, AutomationBench, Terminal-Bench 2.1, and HealthBench Professional.

The clearest signal:

• Compared to Opus 4.8, Fable 5 is the bigger jump for ambitious coding and long-horizon work

• Compared to GPT-5.5, Fable 5 looks stronger across most of Anthropic’s listed benchmark categories, while GPT-5.5 still stays competitive in coding workflows

• Compared to Gemini 3.1 Pro, Fable 5 shows stronger benchmark results in the official table, especially on coding, reasoning, legal, health, and tool-use tasks

The important caveat is that Fable is not “better for everything.” Some benchmark rows include safety or fallback notes, and Claude Mythos Preview is still close or slightly ahead in a couple of rows. But for the kind of work this MonDive is about, building, reasoning, checking, and working across multiple steps, Fable 5 is clearly the model Anthropic is positioning above Opus 4.8 and against the strongest frontier models.

How to Prompt Fable 5

Fable 5 works best when you stop treating it like a one-shot command model. Anthropic’s own prompting guide says it is built for complex, long-running, and ambiguous work, so the prompt needs to help it understand the goal, find missing pieces, and verify the work as it goes.

Prompt you can use:

I’m working on [project] for [audience or goal]. Before building, identify the blind spots, risks, hidden constraints, and key decisions we need to make. Ask only the questions needed to remove ambiguity. Then create a clear implementation plan. After that, build step by step, verify progress against actual outputs, and only report what is confirmed. Do not add extra features or refactor beyond the goal.

The main idea is simple: let Fable think before it builds. The more complex the project, the more value you get from using it as a planner, verifier, and builder instead of just asking it to generate the first version.

Feature comparison: Claude Code vs Codex

One of the best ways to use Fable 5 is to clone software that already exists and customize it for your own workflow.

For this demo, we’ll clone WhisperFlow, a well-known voice-to-text app. The goal is not to rebuild every advanced feature. The goal is to recreate the core workflow locally: speak into your microphone, transcribe the audio, clean up the text, and place the final result into a text box.

This is a good Fable 5 use case because it shows the model handling planning, app logic, local tools, UI, and real workflow behavior in one project.

Step 1: Select Fable 5

Start by opening the Claude Desktop app, then go to Claude Code and select Fable 5 as the model.

But do not start building immediately.

Fable 5 is powerful, but it also burns through usage quickly. So the smarter workflow is to save Fable for the actual complex build and use a cheaper model first to research and prepare the plan.

Step 2: Research the app with Opus 4.8

Before building, switch to Opus 4.8 and use it to research how the app should work.

This saves Fable 5 for the part where it matters most: executing a clear, well-planned build. Instead of giving Fable a vague “clone WhisperFlow” prompt, we first create a proper plan with the core features, local architecture, risks, and success criteria.

Prompt:

I want you to deep research  to come up with a plan to clone Whisperflow. I want you to do some deep research on how Whisperflow (https://wisprflow.ai/)works and what we would need to recreate its base functionality on our computer. Furthermore, I'd like to recreate it locally. So I'd want it to be a local model running on 
Ollama that essentially does what Whisperflow does.

Step 3: Turn the research into a Fable prompt

Once the research is done, ask Opus to turn that report into a clean Fable-ready prompt.

This step matters because Fable works best when the goal and success criteria are clear before it starts building.

Prompt:

Use the full research report you just created above.

Do not browse again.
Do not install anything.
Do not run terminal commands.
Do not modify files.
Do not start coding.
Do not ask me questions.

Your only task is to convert your research report into one final, highly detailed Claude Code /goal prompt that I can paste into Fable 5 for the actual build.

Important context:
I want to build my own local Wispr Flow-style app for Windows.
I do not want you to simply install and configure an existing app as the final solution.
You may use whisper-local, OpenWhispr, whisper.cpp, faster-whisper, Ollama, and similar tools as research references or implementation inspiration, but the Fable 5 task should be to build our own working MVP project.

Do not copy Wispr Flow branding, UI, assets, name, or proprietary design.
The goal is only to recreate the base functionality locally.

The app should do this core loop:

1. User presses a global hotkey.
2. App starts recording microphone audio locally.
3. User presses the hotkey again, releases it, or the app detects silence.
4. App stops recording.
5. App transcribes speech locally using Whisper, faster-whisper, or whisper.cpp.
6. App applies fast rule-based cleanup by default.
7. App optionally sends the raw transcript to a local Ollama model for smarter cleanup and formatting.
8. App copies the cleaned result to the clipboard.
9. App pastes the cleaned result into the currently focused text field.
10. Nothing is sent to any cloud API.

Hardware constraints:
- Windows first.
- CPU-only machine.
- No NVIDIA GPU.
- 16 GB RAM.
- Use CPU-friendly Whisper settings.
- Ollama cleanup may be slow, so the MVP must not depend on Ollama for every dictation.
- Default mode should be fast local dictation with rule-based cleanup.
- Ollama should be optional/on-demand for better formatting.

The final /goal prompt must tell Fable 5 to build the project in phases:

Phase 1:
Create the basic working MVP:
global hotkey → record mic → local Whisper transcription → basic cleanup → paste into active app.

Phase 2:
Add optional Ollama cleanup:
local LLM cleanup, prompt tuning, timeout handling, fallback to rule-based cleanup if Ollama is unavailable or slow.

Phase 3:
Add reliability features:
config file, model selection, hotkey setting, audio device detection, error handling, logs, and clear setup instructions.

Phase 4:
Only after the MVP works, add polish:
tray icon or small floating UI, status indicator, history, custom commands, personal dictionary, and voice shortcuts.

The /goal prompt must include:

1. Clear project goal
2. Product requirements
3. Technical architecture
4. Recommended tech stack
5. Exact folder/file structure
6. Build phases
7. MVP success criteria
8. Testing checklist
9. Error-handling requirements
10. Setup instructions for Windows
11. Performance decisions for CPU-only hardware
12. Instructions to keep the project simple and avoid premature UI polish
13. Instructions to test each major component before moving on
14. Instructions to avoid cloud APIs completely
15. Instructions to create a README with step-by-step setup and usage

Very important:
The Fable 5 prompt should be written as a long-running Claude Code /goal task.
It should be detailed enough that Fable 5 can start building without needing more clarification.
It should strongly instruct Fable 5 to first make the basic dictation loop work before adding advanced features.
It should also warn Fable 5 not to waste usage on unnecessary subagents, dynamic workflows, over-engineering, or large rewrites before the MVP is working.

Output only the final /goal prompt in one markdown code block.

Do not explain it.
Do not add commentary.
Do not start implementation.

Step 4: Switch to Fable 5 and build

Now switch the model back to Fable 5.

Copy the /Goal prompt that Claude created from the research, paste it into Fable 5, and let it run. The point here is simple: Opus prepared the plan, and Fable handles the long build.

If Fable asks a question during the process, answer based on how you want your version of the app to work. Otherwise, let it continue until it reaches the end state.

Step 5: Final result

The final app works like a local WhisperFlow-style voice-to-text tool.

Fable 5 built the full workflow without me writing a single line of code, and the important part is that it runs on my own computer instead of sending everything to a cloud app.

To use it, I placed the cursor on a blank page and held Ctrl + Alt + Z. A small status icon appeared, showing when the app was listening and when it was transcribing.

From there, it captured my voice and placed the cleaned text directly into the page.

The result was simple, but the workflow was the impressive part. Fable 5 handled the full local setup, microphone input, transcription flow, and text insertion without needing me to manually code the app.

This is the bigger lesson: if Fable 5 can clone a paid voice-to-text workflow and make it run locally, you can use the same process to recreate other paid apps or websites, customize them for your own needs, and run them on your own machine.

Fable 5 Use Cases in Action

1. A full Hogwarts-style 3D build

Matt Shumer showed Fable 5 generating an entire Hogwarts-style castle scene with classrooms, the Great Hall, and a Quidditch pitch included. The impressive part is not just the theme, but the scale: this is the kind of world-building task that needs layout, spatial consistency, interior detail, and a connected environment instead of a single static image.

2. Animated, award-style websites

Viktor Oddy’s example shows Fable 5 being used for the kind of website people usually expect from a designer or creative developer. The page had motion, polish, and a more premium feel than a normal AI-generated landing page. This is where Fable starts to look useful for founders, creators, and small teams who want a website that feels designed, not just assembled.

3. A local business automation idea

Everest Chris shared one of the more practical examples: using Fable 5 to find sold homes, identify patios with no shade in extreme heat, and turn that into a postcard outreach idea. That is not just “build me an app.” That is a business workflow: find the signal, create the lead, and automate the next step. That is where Fable 5 starts to feel less like a coding toy and more like a tool for building small business machines.

A Complete Guide to Fable 5

Know someone who may be interested?