Every run of my NYC prompt since April.
24 of them, side by side. Same prompt every time. One HTML file, three.js, no assets, everything procedural. Opus 4.6 gave me boxes on a green plane. GPT-5.5 Standard gave me an empty sky. DeepSeek V4 came back almost black. Then Opus 5
You can use GPT-5.6 Luna and DeepSeek V4 Pro as full coding agents for $0.
They can literally read your repo, write your code, run commands, and build features for you. Freebuff can: → understand your entire repo → write + edit code → run commands → build features for you
GPT5.6 LUNA vs deepseek_ai Deepseek 4 Flash (0731)
10 Websites - Same prompt
7 hours later (6 hours of autonomous execution) I already have the app working on my phone.
I started with DeepSeek and used it for 90% of the run, then switched to Composer 2.5 to take advantage of the speed. $6 from Opencode Go and 28M tokens in Cursor Grok and Composer. It's worth it?:
You have reached the end of the archive
All of deepseek