AI Code Generation Benchmark β built IN NyxCode.
This benchmark page is itself a NyxCode app. The arena.nyx file compiles to the index.html you see deployed. That's the best proof that NyxCode works.
| Scenario | NyxCode | Next.js 15 | FastAPI | Token Savings |
|---|---|---|---|---|
| Blog with Auth | 2,921 tokens / 1 file | 5,589 tokens / 8+ files | 4,820 tokens / 7 files | -48% vs Next.js |
| Todo App | 980 tokens / 1 file | 2,450 tokens / 5+ files | 2,100 tokens / 5 files | -60% vs Next.js |
| Chat (Realtime) | 1,520 tokens / 1 file | 4,200 tokens / 9+ files | 3,600 tokens / 7 files | -64% vs Next.js |
npm i -g @fabudde/nyxcode
nyxcode build arena.nyx
# β dist-site/index.htmlOr with the local compiler:
cd nyxcode-arena
node ../nyxcode/dist/cli.js build arena.nyx
cp dist-site/index.html index.html- Token counting: OpenAI
cl100k_basetokenizer (tiktoken) - Line counting:
wc -lexcluding blank lines and comments - Fair comparison: Identical features in each scenario
- Blog tokens: Measured values from NyxCode v0.12.7 benchmark
- Todo/Chat tokens: Estimated using measured efficiency ratios
Live: fabudde.github.io/nyxcode-arena
One .nyx file = full-stack app with DB, Auth, API, and frontend. Compiles to HTML+CSS+JS (frontend) and Express+SQLite (backend).
npm i -g @fabudde/nyxcode
nyxcode dev app.nyx- npm: @fabudde/nyxcode
- GitHub: fabudde/nyxcode
MIT β Fabian Budde & Nyx π¦