낮밤이 바뀌는 3D 지구본 대시보드

게시물 내용

Kimi K3.1 Pro beats GPT-6 Astra and Fable 5.1 on Terminal-Bench 4.0. But how good is it at actual coding? So I gave it the frontend test I run on every major model release, this time using Kimi Code: One prompt + one reference image. The task: rebuild a 3D globe dashboard. 🌎 Two things stood out: 1. Day/night mode toggle. The reference image shows this toggle, but my prompt never mentions it. K3.1 Pro spotted it and built it anyway. 2. 2D/3D view switching. It built both views, including a working 2D mode. The Three.js globe also looks close to the reference. In night mode, city lights appear and disappear naturally as the globe rotates. K3.1 Pro reads the reference image like a senior frontend dev reading a design spec. It catches details beyond the written instructions. From this test, I'd put K3.1 Pro on par with GPT-6 Astra and Opus 5.5 for frontend and visuals. On a few small details, I liked Kimi's version better.

— @hqmank