Video summary

I Made GPT-6 Astra and Fable 5.1 Build the Same App (RAW RESULTS)

Main summary

Key takeaways

Product Review

Product reviewed (implied)

The video compares AI “app builders” used via desktop chat apps:

  • GPT-6 Astra (via ChatGPT desktop app)
  • Claude Fable 5.1 (via Claude desktop application)

The “product” being effectively reviewed is the ability of these models to generate working, feature-rich Mac desktop apps from the same prompts.

Test setup / user experience (applies to all builds)

  • The same idea is sent as the same query to both models for five builds.
  • Apps are created as native Mac apps.
  • Five parallel sessions are run in each desktop app (5 GPT chats + 5 Claude chats).
  • Working apps are produced without manual debugging:
    • No directing/fixing needed.
    • Only interference occurs when session limits were reached (upgrade required).
  • Speed observation:
    • GPT-6 Astra finished all five projects in ~30–40 minutes total
    • Claude Fable 5.1 completed only one first, then took another ~30–40 minutes for the rest

Unique points mentioned

  • Both models can generate working apps from scratch.
  • GPT delivered faster completion across multiple simultaneous sessions.
  • Claude lagged initially but eventually finished the rest.
  • Minimal user effort during development (except session-limit upgrades).

Build 1: CRM system for a carpet cleaning business

What’s included (in both)

  • A “Today” page with:
    • tasks for today
    • scheduled money
    • team/work organization
    • task cards
    • map-based task tracking
  • Accounts/payment views with convenient:
    • link-through for overdue payments

Winner / comparisons

  • Claude wins overall on usability/design
    • More visually like a real CRM
    • Better-feeling overall layout
  • GPT has gaps / weaker UX
    • Major complaint: no dedicated “Tasks” tab (only appears via a diary-like view)

Key cons noted

  • Claude (Brightway / task booking flow):
    • Booking a new assignment from the main card is clunkier:
      • if a customer doesn’t exist, you can’t create a new customer from the booking card
      • you must go to Clients tab to add, then return
  • GPT (Daymark / CRM):
    • No “Tasks” tab that lists all work conveniently
    • Detail/interaction feels bland vs Claude

Settings limitation (called out as a minus)

  • On Claude’s CRM settings page:
    • Can add business details like “name on the van,” tax number, etc.
    • Cannot add extra vans or employees, despite UI implying multi-van/assignment potential

Verdict for CRM

  • Pros: Claude’s CRM aesthetic + structure; better task-board organization
  • Cons: customer-creation workflow and settings extensibility limitations
  • Overall: Claude is preferred for CRM usability/design, though not perfect

Build 2: Notion clone

Comparison summary

  • Fable 5.1 (Claude) looks more like real Notion:
    • Better alignment of UI proportions and dropdown/button appearance
    • More Notion-like workspace elements (favorites/private/team-space-related differences noted)
    • Slash commands / embed/page creation flows appeared to work
  • GPT-6 Astra (ChatGPT) feels more functional:
    • UI buttons/settings appear to open working settings pages
    • Supports dark mode and functional navigation

Tradeoff conclusion made by narrator

The narrator weighs:

  • “Looks nice but doesn’t work” vs “looks simple but works.”

They argue:

  • Claude optimized for visual similarity
  • GPT optimized for functionality

Also noted:

  • Some Claude buttons looked present but felt useless or non-functional (e.g., “AI button” area behavior)

Verdict for Notion clone

  • GPT wins on practical functionality (everything opens and works)
  • Claude wins on visual resemblance, but with imitation-style UI that can fail to behave like the real product

Build 3: Clothing brand + Shopify-like site (“Knight Index” / raincoat theme)

Claude vs GPT differences

GPT-6 Astra

  • Better image generation (narrator credits built-in image generation)
  • Creates a polished rain-themed brand look (rain/drizzle effect)
  • Store flow includes:
    • product listing + product detail
    • cart + test checkout
    • narrator confirms the test order flow works

Claude Fable 5.1

  • Weather-based pricing mechanic
    • Discount if it’s raining at the user’s coordinates
    • Checked every ~5 minutes
    • In the demo location it appeared “dry,” so no discount
  • Strong image presentation:
    • double-display / hover-like image behavior (front/back or alternate view behavior)
  • Major limitation in commerce flow
    • narrator can’t properly add products to cart / make a fake order
    • cart button not visible / cart behavior missing or incomplete

Model/image tech callout

Claude’s site images were generated using:

  • MM-flux (diffusion models on Apple silicon)
  • Z Image Turbo model

Verdict for clothing brand

  • GPT wins because the store checkout/cart flow works end-to-end
  • Claude impresses visually and with the rain-pricing concept, but shopping/cart checkout is incomplete

Build 4: Interactive 3D model of Melbourne CBD

Claude strengths

  • Provides specific data references:
    • ~16,500 buildings in OpenStreetMap
    • ~1,600 with recorded heights
    • ~40 landmarks
  • Strong interactive features:
    • hover/click building info (addresses/heights)
    • day/night cycle and aerial viewing

GPT-6 Astra strengths

  • Narrator preference for an overall “premium/refined” experience
  • Better night view preference in the demo
  • Smooth flight/zoom navigation with city-tour hints

Issues / dislikes noted

  • Claude’s interface/management described as “bad” at one point (navigation frustration)
  • GPT lacks some automatic shadow progression detail compared to Claude (narrator explicitly notes the downside)

Verdict for 3D Melbourne

  • Closest overall, but narrator calls GPT the clearer winner for refinement/premium feel
  • Claude still impresses with data richness and some day progression behavior

Build 5: Age of Empires II-style game replica (3D version)

GPT-6 Astra vs Claude Fable 5.1

  • Both produce a playable game with similar 3D objects/graphics vibe
  • GPT controls were smoother:
    • zoom in/out felt smooth and responsive
    • plus/minus convenience buttons
    • returning to base was easy
  • Claude controls were problematic:
    • narrator says controls are “terrible” / keys feel mixed (W/S behavior wrong-like)
    • zooming/angle control felt harder

Gameplay/interface comparisons

  • Both allow building elements (houses/farms/barracks) and resource gathering
  • GPT’s interface felt more coherent / like a unified game experience
  • Claude’s management felt confusing:
    • narrator says it doesn’t feel like Age of Empires II in a natural way

Verdict for game

  • Both work, but GPT is preferred due to better controls and game-management feel

Overall conclusion / recommendation (final verdict)

  • The narrator’s final takeaway: GPT-6 Astra is generally the clearer winner across most categories because it more consistently produces:
    • working functionality (cart/checkout, button actions, settings)
    • smoother user experience
    • faster end-to-end completion in the multi-session test
  • Claude Fable 5.1 is strong for:
    • more realistic UI resemblance (notably CRM and Notion look)
    • creative mechanics (weather-based discount concept)
    • good-looking 3D interactions and some map/world features

Final recommendation

  • If your priority is apps that work reliably end-to-end with functional UI: choose GPT-6 Astra.
  • If your priority is closer visual imitation and certain novel interaction concepts: Claude Fable 5.1 can be compelling, but with more functionality gaps noted.

Speakers / perspectives

Only one main speaker/narrator contributes the evaluation throughout the video.

Original video