AutorunBizLABAutorunBiz home ↗
← AutorunBiz Lab

ONE BRIEF · THREE FIRST ATTEMPTS

THE DESIGN WINNER.THE WORKING BUILD.

Sean’s visual favourite was Terra. Sol was the only build that passed every specified functional check.

VANTA MOTOR is fictional. Each model received the same brief, starter kit, fonts, images, tool allowance and high-reasoning request. There was one fresh session per model, no follow-up repairs and no selective reruns.

Blind preference: Terra → Sol → Luna.

Sol was the only full functional pass under the specified checks. Visual preference and working behaviour are reported separately.

OPEN THE RAW BUILDS

LOOK. THEN TRY IT.

These are the uncorrected benchmark entries. Known defects remain visible; no real enquiry or booking is sent by their local forms.

Entry C: 6.1 Terra's VANTA motorcycle concept hero

ENTRY C · First visual preference

6.1 Terra

Observed build: 6m 50s

Known interaction failure

The product image intercepts finish-control clicks, preventing reliable pointer selection and colour changes.

Open 6.1 Terra ↗
Entry B: 6.1 Sol's VANTA motorcycle concept hero

ENTRY B · Second visual preference

6.1 Sol

Observed build: 7m 49s

Passed all specified functional checks

Mobile navigation, finish switching and local demo-form checks passed. This is a tested prototype, not production certification.

Open 6.1 Sol ↗
Entry A: 6.0 Luna's VANTA motorcycle concept hero

ENTRY A · Third visual preference

6.0 Luna

Observed build: 3m 41s

Known mobile failures

Horizontal overflow and mobile navigation failed the common functional checks.

Open 6.0 Luna ↗

THE REVEAL

SAME INPUTS.
DIFFERENT TRADE-OFFS.

The blind entries revealed: A is 6.0 Luna, B is 6.1 Sol and C is 6.1 Terra
First-viewport captures. Open the builds above for the complete pages and mobile layouts.

Recorded previews, not a speed race. This clip is silent.

What this test does—and does not—show

  • One brief and one first-attempt session per model support a result on this task, not a universal model ranking.
  • Observed completion times include agent work and transport overhead. They do not establish intrinsic model speed.
  • Each model could self-check during its original allowance. No changes were made after the runs to improve the websites.
  • The fictional brand, generated imagery and local-only forms are not real product, customer or booking claims.
  • The comparison does not turn a failed raw prototype into a delivery-ready client website.
Back to AI experiments →

Copy project link

Copy this stable public URL. It becomes shareable after the gallery is published.

Open project in this preview ↗