Recently Claude Opus 4.5 has been introduced as the newest “Best Model In The World” :-). Seems like we are getting a new best model every few days, but how much of this is hype and what is, in this case, the truth? Not diving into the benchmarks, I decided to test drive it with a small weekend project. Since I was a game developer in a previous life, and as an aficionado of retro gaming, I decided to try and make an 80s Arcade-like 2D space shooter reminiscent of Galaga, Galaxian, and the like. I wanted to build it with Phaser.JS and not just single prompt a shoddy game or prototype but implement it step by step until a level of polish and playability was reached that is good enough for myself wanting to play it.

So, how did Opus (and Claude Code) perform?

End-to-end implementation with just prompting. I could implement it from beginning to end with just prompting and playtesting. It never felt like Claude Code was getting derailed, it seemed very understanding of functional requirements, technical patterns and even sound and graphics dev, and never worked itself into a typical LLM loop where things spiral far away from a simple and clean solution.

Graphics and sound generation. Claude also did the graphics and sounds, which means I let Claude Code also “draw” all sprites with graphic primitives and play sounds (via Web Audio API). (so no prerendered graphics or sounds)

Handled complexity well. It never got too complex for Claude/Opus to follow, which is a feat because with all the different state machines, layers, animations, UI patterns, retro effects, particle systems, etc… even a simple game like this can get quite complex.

Cross-platform support. I decided to support both mobile devices with touch and desktop devices with mouse / trackpad, resulting in different UI, controls, and other device-related restrictions. This added another layer of complexity, but again, that posed no problems at all!

Full backend in one go. It even built me a Node.JS highscore backend API with game stats, data anomaly detection, database storage in postgresql, etc… This feature was written in one go! Smooth!

DevOps setup included. Opus also helped me to set up GitHub based CI/CD, database migrations, etc… so I can simply (re)deploy by merging to main.

Claude Code’s interactive plan mode. Claude Code’s plan mode now generates and follows editable plan .md files, I have a workflow with a /tempdoc slash command which does sort of the same (writes down plans in markdown and keeps status updates in there), which has now become obsolete :-)

Fun guaranteed. It is very satisfying to turn a gameplay idea into something fun and playable, and finetune and polish until it shines. Recommended! You own the (creative) process, and Claude just does the tedious stuff!

Claude Code is getting better and better! It was lots of fun to do.#

So, is this the best model in the world? I don’t know (and I haven’t played with Gemini Pro yet), but I have the impression that the Claude Code/Opus combo is really getting more and more usable for real world coding scenarios. I only know there is no way I could do this in a weekend without an awesome tool like Claude to speed me up. And to me even more important than the productivity improvement is the satisfaction that comes from being able to focus on the fun stuff, creativity, features, gameplay, instead of writing boilerplate code, looking up examples, etc…

So check it out and Play it here (mobile or desktop), try to break my highscore, and let me know what you think about it.

I am looking forward to building more stuff and finding out all the cool things that other people make with Opus 4.5! What have you done with it so far?