Have you ever stopped to analyze how much of your prompt payload (and API budget) is wasted on just "syntax"? Brackets, curly braces, double quotes, commas, colons, and the endless repetition of keys.
JSON was built for the web, not for LLMs. Language models understand whitespace and indentation perfectly (thanks to Python), yet we insist on feeding them an extremely verbose syntax.
That’s exactly why I built ZEON (Zero-overhead Encoding Object Notation).
In ZEON, we ditch the JSON bureaucracy and adopt a "Tabular Grammar". What used to cost 103 tokens in JSON now costs only 45 tokens in ZEON. That’s a 56% reduction in token consumption, while maintaining perfect readability and structure for AI.
How does it work?
What does this mean in practice?
- More context window available for the prompts that actually matter.
- Smaller bills from OpenAI/Anthropic/Google APIs.
- Faster structured generations and responses (fewer output tokens generated = higher speed).
If you work with LLMs and structured data, check out the repository, drop a star, and download our VS Code extension!
🔗 Repository: https://github.com/Fallen-sch/ZEON
🔗 Playground & Token Comparison: https://zeon-eight.vercel.app
📦 NPM: npm i zeon-format
🐍 PyPI: pip install zeon-format
Top comments (0)