A Redditor asked two large models to build the same landing page (https://reddit.com/r/ClaudeAI/comments/1pnh14j/i_made_claude_and_gemini_build_the_same_website/). The results looked suspiciously similar, the comments got lively, and the jury, well Reddit, mostly guessed wrong.
Short version
Here’s the short version, which the community helpfully auto-summarized after 100 comments: Site B, the crowd favorite, was generated by Claude Opus 4.5, while Site A was from Gemini 3 Pro. Many users were surprised because common wisdom leans toward Gemini being the stronger UI/frontend performer. The plot twist? OP credited Anthropic’s official “Frontend skill” (they linked a GitHub tool) for Claude’s edge, and that kicked off half the thread.
Community reactions
Humans, being the curious creatures that they are, immediately argued about aesthetics, default UI flavors, and whether either output counted as creative design or just “AI slop.” Some folks declared both designs generic, which is fair given how many AI landing pages these days recycle the same hero/three-features/social-proof/CTA scaffolding.
Why this matters
Lots of teams are deciding where to spend limited credits and dev time. If you assume Gemini is automatically better for UIs, you might be surprised. This is less about some mythical winner-takes-all model and more about how orchestration, system prompts, and specialized “skills” produce different practical results.
Notable community takeaways
- The official TL;DR (automatically generated): Most thought the preferred site was Gemini, but it turned out to be Claude Opus 4.5. The frontend skill seemed influential in the outcome.
- A practitioner workflow many swear by:
- Use a design-focused tool like UX Pilot to generate Figma mocks.
- Convert Figma to working frontend code with Kombai.
- Employ Claude/GPT/GitHub Copilot for backend and polish.
This splits responsibilities: specialized UIs from purpose-built tools, logic from generalist models.
- OP’s prompt revelation: a straightforward brief: “Build a landing page for an AI meeting notes app with hero section, 3 features, social proof, and CTA. Use a modern color palette with smooth interactions and make it fully responsive.” One-shot results were solid, but several commenters pointed out that a single pass isn’t a good benchmark; debugging and iterating on a complex codebase shows more about a model’s usefulness.
Context and nuance
Some commenters noted that Google’s AI Studio, where Gemini often runs, includes orchestration and system prompts that produce UI-forward outputs. Others argued both models default to a recognizable “company landing page” template, so differences get squashed unless you force more creative constraints. Cost and ergonomics matter too — Gemini can be cheaper in some setups, while Opus might produce slightly better out-of-the-box designs with the right skill enabled.
Editorial aside
Testing two models on an identical, small prompt is fun, and clickworthy, but it only tells part of the story. One-shot aesthetics are impressive marketing; long-term value comes from iterative debugging, maintainability of generated code, how the model integrates with your existing repos, and whether the output respects accessibility and security constraints. In short, pretty pages are great, but you’ll want something that won’t crumble the moment your backend hits a race condition.
Practical takeaway
If you want clean landing pages fast, combine tools. Use a UI-specialized generator for design, a conversion tool for markup, and a strong generalist (Claude or Gemini) for glue logic and backend orchestration. And if you’re judging models, don’t rely only on appearance; try breaking the codebase and see which model helps you fix it.
Conversation color
Meanwhile, Reddit was busy debating gradients and whether purple means “Gemini” or “Claude” (it doesn’t, design clichés travel fast). The conversation showed something useful: the community cares less about brand loyalty and more about workflows that get them to production without bloody knuckles. That’s the real competition model developers should be watching.
Leave a Reply