Gemini 3.7 Flash vs 3.6: both wrote the same headline
One landing-page brief, one shot each. 3.7 Flash came back self-contained; 3.6 Flash fetched two fonts from Google and scrolls 33 px sideways on a phone. Both opened with the same sentence.
Gemini 3.7 Flash (High)vsGemini 3.6 Flash (High)
No single winner
The checklist goes to 3.7 Flash, 7 of 7 against 5 of 7. The demo goes to 3.6 Flash, and that is not close either.
Gemini
- Technical score
- 100
- Web app completion time
- not recorded
- Requirements met
- 7/7
Gemini
- Technical score
- 71
- Web app completion time
- not recorded
- Requirements met
- 5/7
Word for word, as both models received it
Build a complete modern landing page for an imaginary AI productivity app called "FlowPilot". Requirements: - Use only HTML, CSS, and vanilla JavaScript - Everything must be contained in a single HTML file - Create a premium, modern SaaS aesthetic - Include a hero section, navigation, feature cards, product demo section, testimonials, pricing, FAQ, and footer - Add tasteful animations and micro-interactions - Make the page fully responsive - Include an interactive demo somewhere on the page - Avoid generic gradients and cookie-cutter AI website design - Use realistic copy instead of lorem ipsum - Do not use external libraries or frameworks You have full creative control over the visual direction. Return the complete working HTML file.
Scroll inside the box for the rest of the brief.
Requirements met
| Requirement | Gemini 3.7 Flash (High) | Gemini 3.6 Flash (High) |
|---|---|---|
| No external assets, libraries or CDNs | Met | Not metRequests fonts.googleapis.com, fonts.gstatic.com |
| Nothing scrolls sideways on a 390 px phone | Met | Not met33 px wider than a 390 px window |
| Everything in a single HTML file | Met | Met |
| Vanilla HTML, CSS and JavaScript — no framework in the file | Met | Met |
| Every section the brief listed: nav, hero, features, demo, testimonials, pricing, FAQ, footer | Met | Met |
| Real copy, not lorem ipsum | Met | Met |
| Responsive: viewport set and breakpoints in the CSS | Met | Met |
Automated checks
Gemini 3.7 Flash (High)
Gemini 3.7 Flash (High) · web app
- Total size
- 126.1 KB
- Files produced
- 1
- Outside requests
- None
- Technical score
- 100
Gemini 3.6 Flash (High)
Gemini 3.6 Flash (High) · web app
- Total size
- 143.1 KB
- Files produced
- 1
- Outside requests
- fonts.googleapis.com, fonts.gstatic.comfailed
- Technical score
- 71
Timings are measured in each provider’s consumer web app, so they include network and interface conditions — they are not model inference latency.
What the numbers leave out
Both pages open with the same sentence. Not the same idea — the same words: “Your work moves at the speed of thought.” 3.6 Flash finishes it with “FlowPilot runs the rest”, 3.7 Flash with “FlowPilot orchestrates the rest”, and nothing in the brief asked for a headline, a tone or a phrase. Two builds of one model, given an invented product and full creative control, reached for the same line.
On the requirement list the newer build wins, and it wins on the two things a single-file brief is actually about. 3.7 Flash returned one file that asks the network for nothing. 3.6 Flash's opens by fetching Plus Jakarta Sans and JetBrains Mono from Google — three requests before the headline can paint in the typeface it was designed in — and on a 390 px phone its document is 423 px wide, so the whole page slides sideways under a thumb. The thing pushing it out there is the demo's own control bar, 61 px past the edge of the screen.
Then you press the buttons and the ranking inverts. 3.6 Flash's demo is the better thing by a distance: Trigger Simulation streams a nine-step execution log, relabels itself Re-Run Simulation, and takes about three seconds to play out. 3.7 Flash's Execute Flow re-renders the scenario already on screen — the label flashes “Synthesizing…” and the output panel comes back identical, because the input box still holds the prompt that produced what is there. It does work: pick a different preset, or type an instruction of your own, and it builds a new flow with your words in it. But the first press of the page's primary button changes nothing a visitor can see, and the first press is the one everybody makes.
The brief also said to avoid generic gradients. 3.7 Flash set its headline in cyan to violet to pink — #06b6d4, #8b5cf6, #ec4899, the palette on every AI startup page of the last two years. 3.6 Flash used two stops of blue, and put a theme switcher in the footer so you can have it in purple or green instead.
Neither is finished work. Asked which one we would rather be handed on a Monday: 3.7 Flash's, because a demo that under-reacts is a two-line fix and a page that scrolls sideways on a phone is the kind of thing that survives to production. What this round does not tell you is anything about how long either took — the working times were not recorded — or how they behave on a second attempt, because there was not one.
