I Tested Claude Fable 5 to Build Apps: Surprising Results

Product reviewed

Anthropic Claude “Fable 5” (Cloud model) — described as a “super intelligent” model released alongside Mythos 5, positioned as safer by adding guardrails. The video compares Fable 5 against Claude Opus 4.8 and places it within Claude’s model lineup.

Availability, pricing, and where it fits

• Available in Cloud subscription only until June 22nd.
• After June 22nd: moves to credit-only, with pay-per-use on top of any plan.
• Costs ~2× as much as Opus 4.8.
• Model positioning in Claude lineup:
  • Haiku (smallest/fastest)
  • Sonnet (everyday)
  • Opus (previously top)
  • Fable 5 sits above all
• Claim from the video: Fable beats Gemini, ChatGPT 5.5, and Grok 4 “across pretty much everything” (based on referenced evaluations/benchmarks).

Intended use (per video)

• Built for ambitious, long-running work (not best for quick one-off tasks).
• Suggested strategy:
  • Use Opus/Sonnet for simple tasks.
  • Use Fable 5 for advanced planning where something would otherwise take many prompts (e.g., “20 prompts”).

Core features highlighted (as observed in demos)

• Long-horizon reasoning: could “fill in gaps” from a short prompt.
• Higher complexity execution: more elaborate output in multi-component projects.
• Better animation / richer UI logic in some cases.
• More detailed simulation/game world building: more “lively” and layered.

Demos & outcomes (with comparisons)

1) iOS-style quiz/flashcard app (“Coco Quiz”) — one-shot build

Setup:

• Same short prompt given to Fable 5 and Opus 4.8.
• Used high effort for Fable (not max/extra high) to avoid excessive token burn (max could run for hours).

Observed results:

• Both looked good and were similarly designed at a glance.
• Both had working interactivity:
  • flashcards flip
  • quiz works
  • ability to add items
• Fable 5 added a third feature: a matching game with drag-and-drop and a timer, which Opus didn’t include without being asked.
• Animation quality felt slightly better on Fable’s version.
• Minor difference: Opus included an iPhone frame by default; Fable did not (but could be prompted).

Implied verdict for this app: Fable was slightly ahead due to the added feature and better animation, despite similar baseline quality.

---

2) Web scheduling app (“Calendar” similar to Calendly) — one-shot build

Observed results:

• Both produced a functional single-page responsive app with:
  • booking page (calendar + time slot selection)
  • dashboard/management (reschedule actions, toggles like Monday off/save)
• Design differences:
  • Opus used a purple theme that looked like a “Vibe Coded” style; the reviewer disliked it.
  • Fable produced a different, better-looking theme (subjectively better).
  • Opus had small layout/design issues (e.g., “8 too close to 2.5”; spacing issues in time elements).
• Functionality: worked on both (dummy bookings; reschedule/cancel flows).

Implied verdict for this app: The reviewer doesn’t think Fable is worth the double token cost here. Better results from Opus would likely come from more detailed prompting.

---

3) Responsive landing page (voice dictation site) — one-shot build

Requested sections:

• hero
• testimonials
• pricing (two plans)
• footer
• smooth scrolling + at least one animation

Observed results:

• Both produced complete multi-section responsive landing pages with animations.
• Opus also handled the animation/sections well but used the reviewer’s disliked purple theme.
• Fable seemed more thought-out:
  • testimonials looked nicer
  • FAQ included in the Fable version; Opus appeared to include fewer sections overall
  • section count: Fable had 5 sections, Opus had 3 (per reviewer observation)
• Caveat: Fable’s page had a small formatting issue near the top divider (later looked “fixed” with more context).

Implied verdict for this app:

• The reviewer prefers Fable for design quality, but wouldn’t choose it for a simple landing page without backend work because it would burn tokens faster.

---

4) City simulation game (“SimCity-like”) — stress test

Goal: more realism—buildings, houses, parks/trees, cars/people, traffic lights, traffic jams, and city-life details.

Observed results (Opus vs Fable):

• Both created a city with time controls, zoom, buildings, cars/people, and working traffic light behavior.
• Opus version:
  • buildings looked good
  • cars were mostly box-like with labels
  • people present
  • traffic lights worked
  • night mode turned on building lights
  • cars stopped and formed traffic jams on red
• Fable version (reviewer felt it was clearly better):
  • more elaborate and “busier” city feel (e.g., financial district in the center)
  • cars looked more like cars
  • people looked more like people
  • hovering people gave contextual narration (e.g., “out on a stroll” / “walking off from a long meeting”)
  • traffic lights worked similarly
  • added environmental detail: sun moving and clouds (Opus didn’t)

Implied verdict for this app: Fable was the clear winner—more detailed, more interesting, and more layered simulation.

Main pros mentioned

• Fills in missing details and can add extra features beyond what’s asked (notably quiz app and city simulation).
• Better handling of complex, multi-component tasks (simulation/game especially).
• Richer realism/behavior layers (traffic systems, lights/sun/clouds).
• Helpful for planning and catching overlooked issues (security-minded reasoning mentioned later).

Main cons mentioned

• Cost (~2× Opus): expensive when Opus can deliver similar results.
• For simple sites/landing pages and basic UI, it may be token-inefficient.
• If set too aggressively (max/extra high), a single prompt can run for hours, quickly consuming tokens (hence “high effort” recommended instead).

Reviewer’s “when to use Fable” guidance (explicit recommendations)

Best uses

• Planning/mapping app structure
• Complex projects with many moving parts (e.g., complex desktop apps, CRM systems)
• Security/robustness check before launch (e.g., flagging exposed API keys, DB security gaps)

Likely stick with cheaper models for

• Simple tasks: “landing page or basic website with just text and buttons”
• Cases where Opus can match with better prompt detail

Personal workflow stated

• Use Fable for initial planning
• Hand off to cheaper models for the build once planning is done

Concise overall verdict

Claude Fable 5 is strongest for complex, long-running app planning and multi-component implementations where deeper reasoning and extra layers matter. For simple UI/landing pages or straightforward apps, the 2× cost often isn’t justified versus Opus 4.8—especially if you can get similar results with more detailed prompts.

Recommendation: Use Fable 5 selectively (planning + complex systems + security pre-flight checks), and default to Opus for most day-to-day builds.

Unique points mentioned (consolidated list)

1. Fable 5 released with guardrails; “safer version” of Mythos 5.
2. Fable 5 available in subscription until June 22, then credit-only usage.
3. Costs twice as much as Opus (token/cost emphasis throughout).
4. Model lineup: Haiku < Sonnet < Opus; Fable 5 above all.
5. Claim: Fable beats Gemini, ChatGPT 5.5, Grok 4 across most evaluations.
6. Designed for ambitious, long-running work, not one-off tasks.
7. Suggested use pattern: Opus/Sonnet for simple tasks; Fable for advanced planning that would take many prompts otherwise.
8. Effort level note: use “high effort”; max/extra high can run for hours and obliterate tokens.
9. Quiz app: Fable added a matching game with drag/drop + timer beyond Opus.
10. Quiz app: Fable animations felt better.
11. Quiz app: Opus added an iPhone frame; Fable didn’t by default.
12. Calendar app: both functional; design differences noted (Fable preferred), but cost not worth it.
13. Landing page: Fable avoided the disliked purple theme; added more sections (5 vs 3) including FAQ; minor divider/formatting issue observed.
14. Landing page conclusion: don’t use Fable for simple no-backend landing pages if token cost matters.
15. City simulation: Fable produced a more elaborate, realistic, lively simulation (sun movement, clouds, people narration, more car-like vehicles).
16. City simulation: both had working traffic lights/time/zoom/night lights; Fable was clearly better.
17. Practical guidance: use Fable for complex systems (desktop apps, CRM) and for final security checking (API keys/DB security).

Speakers / perspectives

• Single speaker/reviewer throughout (no other distinct voices introduced in the subtitles).