Fable 5 vs Opus 4.6: I Gave Two Claude Models the Same Prompt - Both Built Me a Space Shooter

I have been using AI coding tools daily at my internship, and one question kept nagging me: how different are two models, really, when you give them the exact same creative brief? Benchmarks tell you about test scores. They don't tell you what it feels like to receive a project from a model - the taste, the architecture instincts, the corners it cuts and the corners it polishes.

So I ran the cleanest experiment I could.

Meet the contenders

This is where the two models revealed completely different personalities. Fable 5 shipped a single HTML document. Everything - markup, styles, the entire game engine - lives in one ~73 KB file. No dependencies, no build, nothing else to download. You could email this game to someone.

Opus 4.6 built a real React application. Vite bundling, componentized code, hashed asset filenames - the architecture you'd expect from a production frontend team. The cost: ~215 KB of JavaScript before styles, plus three font families from Google Fonts, and the page is an empty shell until React boots.

Neither answer is "wrong" - they're answers to different questions. Fable optimized for shipping a finished thing; Opus optimized for a codebase you'd keep working in. For a one-shot prompt with no follow-ups, I have to admit the single file is the sharper read of the assignment.

Fable 5 simply packed more systems into the same prompt - difficulty economy, four power-ups, a voice channel, WPM. Opus went narrower but smoothed the player journey: you're onboarded, you launch a mission, you fail dramatically ("MISSION OVER"), you retry. Fable built a game with more knobs; Opus built a game with better flow.

Here's a detail most comparisons would miss. When I hit both URLs with a crawler (the same way Google or a link-preview bot does), AstroType returned its entire game UI as readable HTML - title, modes, leaderboard, controls, everything. Cosmic Typer returned a page title and an empty <div id="root"></div> - the content only exists after JavaScript runs.

For a game it's a small thing. But scale this instinct to a marketing site or a blog and it's the difference between ranking and being invisible. One-shot Fable defaulted to the crawlable answer; one-shot Opus defaulted to the SPA answer. Worth knowing before you pick a model for a landing page.

Typography is the clearest example of the taste gap running the other way. Opus 4.6 pulled in Orbitron for headings, Rajdhani for UI and Space Mono for the typing surface - a deliberate, sci-fi-coherent type system. It also sweats the feedback details: shields don't just work, the game shouts "Shield Absorbed!"; explosions have blastwave and shockwave effects; the menu says "Type to destroy asteroids..." right in the input.

Fable's presentation is tight and readable, and shipping three independent audio channels in a single file is genuinely impressive - but its visual identity leans more on system defaults. If I were judging pure art direction from a cold start, Opus takes this round.

The scorecard

What I took away as a developer

• Same prompt ≠ same product. Two frontier models converged on the concept but diverged completely on architecture - the "how" is where model personality lives.
• Constraints reveal taste. With no follow-ups allowed, every default matters: rendering strategy, dependencies, fonts, onboarding. You're not just picking output quality, you're picking defaults.
• Judge AI code like you'd judge a teammate's PR. Payload size, crawlability and player experience told me more than any benchmark number could.
• This is the skill now. Writing one good brief and critically evaluating what comes back - that's what working with AI models actually looks like in 2026.

Try both games above, form your own verdict - and if you disagree with mine, I'd genuinely love to hear why. This kind of experiment is exactly how I sharpen my instincts for AI-assisted development, and I'll be running more of them.