Anthropic's Claude Opus 5.5 is turning into the quiet workhorse of the AI world, and the most interesting proof isn't a benchmark chart — it's the odd, useful things people are building with it in their spare time. A roundup on Habr, posted by alexsyntx and already read by a sizable audience, collects eight non-obvious ways to use the model. The author had been running Opus 5.5 daily for more than a week and came away impressed for a practical reason: almost everything they asked for came back nearly error-free, with no need to sit there correcting it. Token costs also landed lower than Fable 5.1, which is the kind of detail that only shows up after you've paid a few bills.
What happened
Here's the twist that makes the whole story interesting. Opus 5.5 cannot generate images or video. It's a text-and-code model. And yet the examples people keep sharing on X are animations, interactive apps, brand decks and playable games — visual things, built with nothing but prompts and JavaScript.
Think of it like a flipbook artist who never touches a pen. Instead of drawing each page, the model writes instructions that tell the browser how to draw each page, then strings 28 seconds' worth of them together. That's the trick behind everything below.
A 28-second cartoon made entirely from code
Creator Kevin Ngo asked Opus 5.5 for a 28-second animated story in which every frame is generated by JavaScript. What came back looks hand-drawn: a city under a starry sky, paper airplanes, a smiling star-shaped creature, a girl writing at a window, and answers that arrive in the form of dogs and tea. The look is paper cut-out, and the motion is code. Another creator pushed the same idea further, animating his whole life from birth to the present with JavaScript brushstrokes — no image model, no video model, no source footage.
A camera lens you can twist with your mouse — for $25.66
Ryan Sal built something more practical: an interactive lens lab that teaches how camera focusing actually works. Turn the focus ring and the glass elements inside the lens shift; the sharp band of focus slides across the scene. Suddenly you understand why photographers talk about depth of field. It was generated in a single run lasting 1 hour 26 minutes, and the API bill came to $25.66. Expensive? The Habr author doesn't think so — he estimates the same app built on Fable 5.1 or GPT 6 Astra would have run into the hundreds of dollars.
Brand decks that look designed rather than generated
Presentation builders like Gamma, Plus AI and NotebookLM have owned this niche, but Opus 5.5 makes a serious case for itself. A deck shared by Yahyavision holds a single futuristic look throughout: a logo built from interwoven elements, the tagline “Every AI Agent, One Index,” website mockups, app icons, and a turquoise-and-blue palette threaded with electric lime and azure.
Games you can actually play
Chetan Ankola posted an updated version of his driving simulator, rebuilt in Three.js with Claude Opus 5.5. Three.js is a free JavaScript library that draws 3D scenes inside a browser tab — no game engine install required. In his sim you steer a car down whimsical, hand-drawn winding roads, past floating islands and strange set pieces with names like Lemon Spiral and Möbius Arc.
What it means for you
You don't need to be a developer to benefit here, and you don't need to care about any of the jargon. The pattern is simple: you describe the thing you want in plain English, and the model writes a small, self-contained file that your browser runs.
At home
Ever tried to explain how a camera's aperture works to a teenager and given up halfway? Ask for an interactive version instead. Anything you'd normally sketch on a napkin — a budget tracker, a chore wheel, a seating plan — can exist as something you click and drag within a few prompts.
At work
The lens lab is the template. Any process your team keeps mis-explaining in meetings can become a one-hour interactive explainer: how the pricing tiers actually differ, where the onboarding funnel leaks, how the quarterly bonus math works. People remember what they can play with.
For business owners
The Yahyavision deck is the interesting case, because presentation tools are one of the most crowded categories in AI. If a single prompt gets you a consistent visual identity — one logo, one palette, one tone across every slide — you just deleted a freelance invoice.
For study
This is where the model quietly shines. Ask for a music theory app where you drag notes around and hear the chord change, or a sorting visualizer you can pause and restart with a fresh shuffle. Both are ideas the Habr author wants to see more of, and both turn a dry textbook concept into something you poke at until it clicks.
For creative side projects and extra income
Animation without an animation tool is a genuinely new capability. If you make content, you can now produce stylized clips that cost nothing but tokens — and the same skill sells as freelance work, since most clients can't tell whether frames came from a render farm or from code. If you want to warm up your prompting before you spend anything, mykreatool.com has free AI tools worth playing with.
How to try it right now
Free option first: open the Claude web app in your browser and pick Claude Opus 5.5 from the model menu. No API key, no code editor, no extra signups. Then:
1. Ask for one self-contained file. A prompt like “Create a single HTML file with inline JavaScript and CSS that lets me drag notes and hear the chord change” works far better than a vague request. One file means nothing to install.
2. Describe the look, not just the function. “Paper cut-out style, muted colors, hand-drawn feel” is what turns a functional demo into something people share.
3. Set a hard constraint. Name the frame count, the duration (28 seconds worked for Kevin Ngo), or the resolution. Constraints are what stop the model from wandering.
4. Save and open it. Copy the output into a plain text file, save it as something.html, and double-click. Your browser does the rest.
5. Only then consider the API. If you want to generate at scale or tweak over and over, the API route is what produced that $25.66 lens lab in 1 hour 26 minutes.
Upsides and what changes
The headline change is that visuals are no longer gated behind image and video models. Code becomes the rendering layer, and code is something a text model is very good at. That flips the cost curve: instead of paying per image, you pay once for a program that can draw anything within its rules.
The second change is reliability. The Habr author's real praise isn't about brilliance — it's that Opus 5.5 needed almost no supervision. For anyone who has spent an afternoon fixing a chatbot's mistakes, that's the upgrade that actually saves hours. Cheaper tokens than Fable 5.1 make long, iterative jobs practical instead of painful.
Limitations
Be clear-eyed about what this isn't. Opus 5.5 doesn't generate images or video, full stop — every visual you see is code running in a browser, which means it only works where a browser (or a compatible runtime) can run it, and it won't hand you a ready-made MP4 file. The prices in this roundup are specific to those builds: $25.66 and 1 hour 26 minutes for one app, through the API, with no guarantee your prompt lands on the first try. The examples are one-off posts shared on social media, so treat them as demonstrations, not benchmarks. Your results depend heavily on how precisely you describe what you want, and a vague prompt will get you a vague, generic result every single time.
Conclusion
The most useful thing about these eight examples isn't any single one of them — it's the underlying move. When you stop asking an AI to write text and start asking it to write a small program, you get things you can click, test and share, and you get them from a plain English description.
Your one action today: open Claude Opus 5.5 and ask it for a single self-contained HTML file that explains something you already understand well — a sorting visualizer, a chord player, a focus simulator. Don't aim for perfect. Just watch your idea turn into something you can touch.
Comments 0