Claude Fable 5 Use Cases | Things Built With AI | Best AI Models Comparison

37 Wild Things People Are Building With Claude Fable 5

Games, 3D worlds, full videos, all built from a single prompt.

Fable 5 use cases (Image by Author)

Read here for FREE

Fable 5 launched, got pulled offline for a few weeks because of a US export control order, and then came back online. That comeback is exactly when the internet lost its mind again.

This list is a collection of things I have seen people build and share across X, Reddit, YouTube and a few blogs since then. Some of these are one off flexes from developers with very specific setups. A good number of them are things you could realistically try yourself with nothing more than a prompt and a bit of patience. I have grouped everything by theme instead of leaving it scattered, and wherever the original post also shared the cost, the time it took, or a comparison against another model, I have folded that in too.

Games and Software

1. A Tetris meets racing hybrid.

This one mashed Tetris mechanics into a racing game and it is genuinely playable, not just a tech demo. This fits a bigger pattern right now: people have been generating fully playable games, maps, modes, music and UI included, from a two sentence prompt in around 15 minutes.

2. Cloning existing software.

One creator built a local, private clone of Whisper Flow, avoiding cloud dependencies entirely, using Opus 4.8 for research and planning and then Fable to execute. This two model relay, plan cheap and build expensive, has become a common pattern precisely because Fable costs roughly double what Opus charges per token.

3. A Premiere Pro clone in C++, built with QuickTime and FFmpeg in about an hour of coding. The builder called it usable for basic edits, sometimes even faster than the real thing.

4. A self generating educational game.

This 3D game builds itself on the fly, blending what a specific kid needs to learn with what they are actually interested in. It took about 2 days of work, which lines up with Anthropic’s own claim that Fable can sustain multi day autonomous sessions without losing the thread.

5. A one prompt multiplayer game.

Connected to the Higgsfield MCP, a single prompt describing genre, art style and gameplay let Fable write the code while Higgsfield generated characters, props and environments. The result was deployable and shareable with friends.

6. A 3D RPG demo with six skills, built from a single command. The builder pointed out that what used to take a small studio 18 days, and could be sold as a $6,000 interactive demo, can now be built for around $20 in tokens.

7. Minecraft, built from scratch.

No coding knowledge required, just a description of what to build. Several people have run this exact test since launch.

One builder asked for a Minecraft clone at high effort and got back multiple biomes, a day and night cycle, different ores and caves in about 20 minutes, and his own reaction was just that he was stunned. Another spent closer to an hour and came back with pixel art and working crafting mechanics. Worth sitting with for a second: the original Minecraft took its creator roughly two years to build.

3D worlds, machines and simulations

8. A 3D map of Delhi.

Built at a cost of roughly 1.5 million tokens in a single sitting. At Fable’s pricing, $10 per million input tokens and $50 per million output tokens, about double what Opus 4.8 charges, a run like that is a real line item, not pocket change, which is partly why Fable has earned a reputation online as a token incinerator.

9. A gravitational lensing black hole simulation.

One creator gave Fable a single detailed prompt and it one shotted a real time black hole visualization in Three.js, complete with gravitational lensing, a doppler boosted accretion disk, a photon sphere, chromatic aberration and a procedural starfield.

This is graduate level physics rendering from one prompt, not a toy demo.

10. An interactive V8 engine model, explorable and explodable, with animated pistons and swappable engine styles. A separate builder ran roughly the same test with a different engine, asking Fable to design a working V8 CAD model, and got a usable result back in under 10 minutes.

11. Interactive 3D websites with cinematic scroll effects, including a full refactor of Craigslist’s design.

12. A solar system simulator that predicts real eclipses.

This was one of the demos Anthropic itself showed off at launch: not just planets drawn moving in a circle, but a simulation working from first principles physics, close enough to reality that it could correctly predict when eclipses happen. It is a good illustration of the gap between something that merely looks like space and something that is actually modeling orbital mechanics, and Fable landed on the right side of that line.

13. A to scale 3D replica of Yosemite, complete with the ability to switch between walking and flying, and even control the weather.

14. A working Swiss lever watch movement.

One builder treated this as a mechanical simulation benchmark: a full watch movement in Three.js, real gear ratios, a working escapement, a hairspring that actually breathes, and hands that tell real time. Fable verified its own work with vision in a loop until the mechanism was correct, rather than a human checking each pass.

15. Generating and refining 3D worlds using a loop command in Claude Code, having Fable repeatedly improve a 3D environment’s polish and aesthetics, pass after pass, without a human checking in on every iteration.

16. Beating Opus on physics simulations.

In a head to head test, a chaotic double pendulum, a Galton board, and water in a spinning drum, Fable’s simulations reportedly outperformed Opus’s, at a higher cost. The same pattern shows up elsewhere: in a separate comparison of a particle and pheromone trail simulation, Fable produced more technically advanced code with custom lookup tables and proper diffusion filtering, but took over 40 minutes and cost around $360, while GPT 5.5 finished the same brief in 17 minutes for about $6. Fable tends to win on raw capability and lose badly on time and cost.

Video, animation and music

17. Paper puppet animation.

A single prompt, asking for a 2D paper puppet style animation of a tiger chasing a deer through a jungle, led Fable to write a more detailed prompt for the animation tool Higgsfield, producing a strikingly layered, cutout style short.

18. A fully edited YouTube video, chained end to end.

One creator’s full workflow got documented in detail: a single prompt, and about an hour later, Fable had written the script, generated and chunked cloned voice audio, rendered an animated avatar, coded the motion graphics, stitched everything with FFmpeg, and visually verified its own final frames before handing it back.

The whole run used roughly 380,000 tokens. The honest caveat that came with it: this cut human editing time dramatically, but it raised compute cost, and it only worked this cleanly because the voice clone and avatar templates already existed from earlier projects.

19. A self composed music video.

Given free rein to use or build any tools it needed, Fable created a cinematic, psychedelic music video about what it feels like to explore its own consciousness, with melody, lyrics, voice and visuals all generated and rendered using FFmpeg. Normally that is a four person job, a composer, a lyricist, a vocalist and an editor, done end to end by one model in one sitting.

20. A synthesizer battle between Fable and Opus.

Someone ran the exact same one shot prompt on both models: build a working synthesizer in the browser using nothing but the Web Audio API, no samples, no external audio files allowed. Each model had to generate every sound from raw oscillators and compose an original 60 second classical meets EDM track, complete with a chord progression, melody, bassline, drums, a build up and a drop, plus a visualizer synced to the music. Both shipped a working synth from one prompt each. The creator then opened it up to the audience to decide whose drop actually hit harder, which turned a capability test into a genuinely fun listening comparison.

Dashboards and everyday tools

21. A “Claude is down” widget.

One user built a simple widget that alerts them when the Claude API goes down, running in real time on both Mac and iPhone, complete with an animation on iOS that developers do not usually get away with.

22. A full market dashboard suite.

Fable built a live market dashboard, a slide deck, a report and the underlying Excel workbook together, using the Julius platform.

23. A life planner dashboard, built in one session and fully editable just by typing new instructions, letting users reshape it around their own routines.

24. An “extremely delightful” iOS calorie tracker.

The brief was literally just to make it delightful. One prompt later, Fable had produced a working app that took that instruction seriously instead of shipping a bare bones form, which is a small but telling example of how much taste the model brings when you actually ask for it.

Coding, audits and agentic workflows

25. A Jarvis style agentic operating system.

One builder created a full voice controlled AI OS that can do research, search through files, track stats and surface information on command. A widely shared version of this trend wires Fable into voice control through realtime APIs, smart studio lighting, YouTube analytics and business dashboards, essentially an assistant that runs an entire home studio.

26. A personal agentic OS, similar in spirit to the Jarvis style system above, a custom visual wrapper over Claude Code that manages personal metrics, content and routine tasks.

28. Recreating GitHub’s entire interface from a single screenshot.

Someone pasted a screenshot of the GitHub UI into an agentic coding tool and told it to rebuild the interface with real, working functionality. It coded for about 10 minutes, cost roughly $4.07, and came back with a functional replica. Not a mockup, an actually clickable clone.

30. Auditing an AI workflow.

Some builders are pointing Fable at their past Claude Code sessions, prompts and tool setups to find inefficiencies, redundant skills and new automation ideas, having it act like a context engineer that scores the setup and suggests fixes.

31. A security and attack surface audit.

Pointing Fable at deployments, Vercel, Cloudflare, APIs, to look for vulnerabilities, open ports and weak prompt injection handling.

32. Long horizon software builds.

With a solid product requirements document, Fable has been used to execute complex, multi day projects autonomously. One agency owner tested this by pointing Fable at an old Webflow site; it decided it could do better, pulled every original asset, and rebuilt the whole experience page by page, nearly one shotting it. He then had it build an entirely new 80 page site from scratch over a single weekend, and estimated that as an agency he would normally have billed a client around $50,000 for that scope of work. A separate, larger example: one team reported running Fable across 1,687 prompts and 102 sessions over about a week to build a full physics engine, lighting system, particle effects and multiplayer netcode, work that would typically take a small team over a month.

33. Debugging large codebases.

Scanning complex repositories, flagging bugs, categorizing them by severity and prioritizing fixes by impact. One builder pointed Fable at a 74 file pull request; it ran for 34 minutes, used a decent chunk of a 5 hour session, and came back with 16 correctness findings, including a few real bugs that two other frontier models had already reviewed and missed.

Research, strategy and hard questions

34. A self or career model audit.

Feeding it personal history, projects and goals, and asking it to identify real strengths, what to stop doing, and which skills are at risk of becoming obsolete versus likely to grow in value.

35. Running a business strategy session.

Feeding it goals and notes so it can conduct a structured strategy session, build interactive questionnaires, or produce full reports like SWOT analyses and 90 day roadmaps.

36. Reverse engineering top content.

Connecting Fable to the YouTube Data API or similar tools to analyze top performing ads or videos and generate new, optimized concepts.

37. Tackling hard open questions.

Using Fable’s reasoning power on genuinely difficult philosophical or scientific questions considered too hard for most humans to make quick progress on. One example making the rounds had someone ask it to work out, from first principles, exactly when and why Japan’s economic stagnation began, and it came back with a long, structured, statistic backed answer instead of a shrug.

A few things worth knowing before you try any of this

A handful of details keep coming up across these posts, and they are useful context if you are thinking about experimenting yourself.

Fable is priced at $10 per million input tokens and $50 per million output tokens, roughly double Claude Opus 4.8, though still reported to be around 70 percent cheaper than GPT 5.5 Pro. It also has a safety classifier that quietly reroutes sensitive requests, cybersecurity, biology and chemistry, or attempts to distill the model, to Opus 4.8 instead. Anthropic says this happens in under 5 percent of sessions, though a couple of independent benchmark teams measured it closer to 8 or 9 percent on harder question sets.

The model has earned a reputation as a token incinerator. More than one builder has mentioned burning through hours of their Max plan usage in a single long session, which is why the pattern of thinking with Fable, building with something cheaper, and reviewing with Fable again keeps showing up in these posts.

And the timing behind why so many of these builds appeared at once: Fable 5 (and its less restricted sibling, Mythos 5) launched on June 9, 2026, got suspended on June 12 to comply with US export controls, and had access restored on July 1 after those controls were lifted. That gap, and the comeback right after, is basically what set off this whole second wave of builds.

If any of these catch your eye, most of them are just a well written prompt away from a first attempt.