Hands-on test of MiniMax M3.1 Flash and Claude Opus 5.5 writing a Three.js solar system

Holy cat, I'm genuinely amazed—this is probably the most shocking and counterintuitive AI large-model hands-on test I've done recently. Right now the whole internet treats Claude Opus 5.5 as the ceiling for coding, so I threw the same Three.js prompt at it and at MiniMax's newly released M3.1 Flash at the same time: build a particle nebula and planetary halos, and when you click a star, the camera smoothly flies to it.

On one side is a closed-source top-tier flagship billed by the dollar; on the other is a domestic speed model that costs only $0.6 per million output tokens and just $0.015 for cache reads. Both ran through on the first try, which really surprised me. But bros, what I find even more outrageous is the much cheaper domestic speed warrior on the left, M3.1 Flash: during code initialization, it detected 4 potential rendering errors on its own, and without anyone prompting it, it ran tests in the background, fixed them on the spot, and locked the frame rate firmly at 60 fps. The key thing is, my prompt didn't ask for a control panel at all, yet it proactively whipped up an extra one in the top-right corner, with one-click toggles for orbit lines, glow, and the nebula—what it delivered was a complete, out-of-the-box work. That's genuinely pretty badass.

Opus 5.5 on the right, meanwhile, held down the aesthetic foundation: the solar halo has a strong cinematic feel, and the layout of the bottom nav bar and the Jupiter and Neptune info cards is extremely refined—in lighting and design taste, the veteran flagship Claude still has its own unique finesse.

But lay both bills on the table, and no developer can avoid this cruel arithmetic: in the vast majority of everyday front-end interactions, code builds, and agent workflows, the capability gap between the two no longer comes anywhere close to matching the price gap between them.

Back in the day, when we tackled complex projects, we always felt we had to keep the most expensive brain attached the whole time.

English text is being prepared.

Still, impressive as it is, we have to look at the benchmarks objectively: it took first place overall in Automation-Bench (62.2), but on the hardcore Terminal Bench 4.0 it scored only 38.1, still a real generational gap from Opus 5's 51.8.

Chewing through a 100,000-line legacy codebase still requires calling in the Claude flagship, haha, but for 70-80% of everyday prototype development and automation scripts, it can already take over steadily—and that alone is enough to save teams a ton of real money, hh.

Honestly, even now I'm still full of feelings about it: domestic models are iterating faster and faster, and getting stronger and stronger. When a fast model that costs only $0.6 per million output tokens learns to run its own tests and debug itself to 60 fps, the premium wall that foreign top-tier flagships built on one-shot intelligence has already started to leak and crack.

After seeing this 3D solar system, what moves you more: the geeky style on the left with full functionality and self-debugging, or the more cinematic design style on the right? Bros, feel free to interact and share your thoughts.

Anyway, if I had to pay out of my own pocket to run batch tasks, for a contender on the left that's fast, cheap, and worry-free, I can no longer find a reason to refuse, hahaha.