Video summary

I Made GPT-6 Astra and Fable 5.1 Build the Same App (It's not even close)

Main summary

Key takeaways

Product Review

Product(s) reviewed

Two AI coding/model options are compared:

  • GPT-6 Astra
  • Fable 5.1 (uses extra high effort in at least some tests)

The video compares both across multiple app/build tasks and tracks token cost using API pricing estimates.

Key features evaluated (by task)

1) E-commerce website (ski clothing brand, heavy asset generation)

Prompt intent: Minimal design direction; generate brand/colors, products, and product images + human models (via Hicksfield). Include a video hero and interactive elements (e.g., 3JS).

Fable 5.1 results

  • Produced a full ecommerce flow: hero video, product catalog, product interactions, and a checkout page
  • Added extra content/sections; includes a 3JS interactive element
  • Minor issues:
    • Background/theme not consistent across pages
    • At least one model has face cut off

Astra results

  • Similar ecommerce structure and checkout flow
  • Feels cleaner/more “premium”; more consistent studio-style backdrop
  • Generates a size guide
  • Overall fewer/leaner sections than Fable, but still looks solid

Cost + time

  • Fable 5.1: ~10 min, $4.70
  • Astra: ~23 min, >2× time and more cost

Verdict for this build: Fable slightly wins on output/design, with cost consideration noted.

2) 3D Kyoto trip map/simulation (Sim City-like use case)

Prompt intent: Rebuild Kyoto in 3D with navigation/zoom/pan, point-to-place interactions, lighting modes, measuring distances, and a “flight tour”.

Fable 5.1 results

  • Very robust 3D experience: zoom, pan, rotate viewpoints
  • Left sidebar of attractions; teleport to places like Kyoto Tower and Yasaka Shrine
  • Includes lighting modes (morning/day/golden hour/night)
  • Includes flight mode (tour) + pause/skip
  • Includes measuring tool (example shown: 2.8 km / 35 min on foot)
  • Not full-city (only the attraction-dense portion), but still detailed
  • User impression: night looked cooler than golden hour

Astra results

  • Smaller coverage than Fable but still detailed
  • Default aesthetic: green color palette (called out as a tendency if you don’t specify colors)
  • Includes measurement and navigation features
  • Adds extra scene detail like a sculpture and visible stone/steps detail when zoomed in
  • Mentions “wireframe/clay” style visualization options

Cost + time

  • Fable 5.1: ~50 min, $146.80
  • Astra: ~40 min, $20

Verdict for this build: Astra wins on efficiency/cost, while Fable wins on larger/bigger-city detail.

3) Clone a product landing page (Mac voice dictation app; reference: Coold do UI)

Prompt intent: Provide reference URLs + adapt copy to “Whisper”; generate motion graphics/reveals/tutorial video and mimic layout/topography.

Fable 5.1 results

  • Strong replication of reference structure and animations
  • Hero landscape with motion; interactive stone reveal with crawling plant
  • Generates tutorial video and many sections (features, screens, testimonials, pricing, FAQ, etc.)

Astra results

  • Hero landscape and motion felt smoother, especially the plant/cursor-following reveal
  • Tutorial video includes motion graphics; user notes it produced this without extra instructions
  • Cohesive and “cleaner” overall presentation
  • Differences/possible issues:
    • Tends toward Astra’s green palette (user says it’s fine/easily changeable)
    • Pricing area seems off: produced a subscription/waitlist-style box instead of clearly matching the reference pricing section
    • Some design differences vs reference; still includes FAQ and a strong footer

Cost + time

  • Fable 5.1: ~35 min, $48
  • Astra: ~~60 min (“about an hour”), $57

Verdict for this build: Astra wins on visual cohesion/motion quality, with slight cost/time penalty.

4) Mobile app end-to-end: receipt scanner + landing page (Expo Go)

Prompt intent: Build full receipt scanning app (scan receipt → extract items → categorize → connect to accounting via LLM API). Uses Claude API for extraction. Also generate app screenshots for App Store readiness. Design: “Notion-style” branding + illustrations; produce UI ready for publishing.

Fable 5.1 results

  • Functional app: receipts list, categories, scan receipt via camera
  • Claude key enabled extraction + categorization; edits supported (currency, items, quantities)
  • UX described as functional but design described as not amazing (more utilitarian)
  • Landing page: generally strong, but Astra’s landing page is later preferred for cohesiveness

Astra results

  • App UI described as cleaner/premium
  • App has navigation tabs with haptic feedback
  • Screenshot count: produced fewer screenshots than Fable (later scenario: 4 vs Fable’s 6–7)
  • Landing page: user later prefers Astra’s app design; landing-page comparison varies across iterations

Cost + time (from the receipt-scanner section)

  • Fable 5.1: ~36 min, $86
  • Astra: ~49 min, $59

Verdict: Split decision

  • Astra preferred for app UI/design
  • Fable preferred for landing page effort/quality
  • Astra missing an iPhone mockup detail despite “iOS app” prompt (user says prompting should fix it)

Main pros and cons mentioned

Astra — Pros

  • More efficient/cost-effective for 3D modeling tasks (Kyoto notably cheaper)
  • Produces smoother motion graphics and cursor-following reveals
  • Tends to deliver cleaner/more cohesive UI (especially app screens)
  • Adds useful 3D features and scene details (sculpture, stone steps, measuring/navigation)
  • Closing implication: faster for “computer use” automation/iterations (relative to Claude-based workflows)

Astra — Cons

  • Tendency to default to a green palette when no design preference is given
  • Sometimes mismatches expected specifics (e.g., pricing section/waitlist confusion, not always using an iPhone mockup when iOS is requested)
  • In some cases may generate less content or smaller scene coverage than Fable (Kyoto scale)

Fable 5.1 — Pros

  • Strong ecommerce completeness and broader content sections
  • In Kyoto, built a bigger/more expansive city portion with impressive interactivity
  • Landing pages can look like they were “spent more time on”
  • Functional app behavior validated: receipt scanning + Claude-based extraction works

Fable 5.1 — Cons

  • Higher cost for complex builds (Kyoto especially)
  • Motion/video work can be weaker: creator notes Fable “struggled” cutting mistakes/pauses; plans to test Astra for video editing next
  • UI/design sometimes less “premium” on first iteration (receipt app described as functional but not amazing visually)
  • Background consistency issues in ecommerce case

Comparisons highlighted

  • E-commerce: Fable wins on “overall output” and score (Astra looks cleaner/premium)
  • Kyoto 3D: Fable wins on scale/detail; Astra wins on cost/efficiency while still delivering excellent usability
  • Landing page clone (Coold do → Whisper): Astra wins on motion quality and cohesion; Fable matches structure but feels less refined in animation smoothing
  • Receipt scanner app: Astra wins on app UI/design; Fable wins more on landing page polish and screenshot richness

Unique points mentioned across the video (consolidated)

  1. Astra released shortly after Fable’s 5 announcement; testing occurred over a weekend
  2. Both can generate full websites/apps with interactive elements and asset generation via Hicksfield
  3. Ecommerce test: human models required; Fable had minor face cutoff; Astra had cleaner studio-consistent backgrounds
  4. Ecommerce checkout pages generated by both; Astra and Fable offer size selection and cart/checkout flows
  5. Kyoto test: both generate interactive 3D with zoom/pan and teleport to landmarks
  6. Fable includes flight mode/tour + measuring tool; distance example 2.8 km / 35 min
  7. Astra includes style toggles (clay/wireframe) and measurement; adds sculpture detail
  8. Kyoto lighting modes included; Fable user preferred night over golden hour
  9. Landing page clone: Astra’s reveal animation smoother; tutorial motion graphics appeared without extra instructions
  10. Astra pricing area mismatched (subscription/waitlist confusion)
  11. Receipt scanner: both created camera scan flows and editable extraction results using Claude API
  12. Astra app UI described as cleaner/premium; includes haptic feedback
  13. Screenshot output differences: Astra fewer screenshots in one case
  14. Astra sometimes misses expected iPhone mockup even with iOS prompts
  15. Final guidance: choose model based on task—Astra for 3D/computer-use/iteration speed; Fable leaning for design workflows and landing pages

Speakers/views

  • Primary narrator (single speaker): runs all tests and provides the comparative judgments (no separate speaker viewpoints explicitly identified)

Overall verdict / recommendation

  • Don’t pick one universally. The video’s conclusion is task-based:

Choose Astra if you need

  • 3D modeling/computer-use efficiency
  • Smoother motion graphics
  • Cleaner app UI

Choose Fable 5.1 if you need

  • More polished landing pages
  • Stronger creative design output
  • Larger-scale 3D city coverage
  • (with the caveat that costs can be much higher and video refinement may be weaker)

In short: Astra is better for motion + 3D efficiency; Fable is better for certain design/landing-page quality and larger-scale scene builds.

Original video