Can a New Language Make LLMs Write Code with 100% Accuracy? Meet Sui

Overview

Sui is an open‑source programming language created to eliminate syntax and spelling errors in code generated by large language models (LLMs). Its core claim is that, once an LLM learns Sui, token consumption becomes more efficient than with conventional languages, potentially enabling 100 % correct code generation.

Design Principles

Zero syntax‑error rate : the language grammar is constructed so that any program written in Sui is syntactically valid.

Zero spelling error : identifiers are numeric (e.g., v01) rather than textual names, removing misspelling possibilities.

Line‑level independence : each line is a self‑contained statement; there is no implicit cross‑line dependency.

Pure logic language : Sui expresses only computational logic. User interfaces can be built separately with any front‑end framework (React, Vue, Hono.js, native JavaScript, etc.).

Token‑efficiency optimization : because the language uses compact numeric tokens and line‑level isolation, an LLM that has been fine‑tuned on Sui can generate code with fewer tokens than when using mainstream languages.

Installation

From PyPI (basic): pip install sui-lang From PyPI with WebAssembly support: pip install sui-lang[wasm] Via Homebrew on macOS/Linux:

brew tap TakatoHonda/sui
brew install sui-lang

Clone the source repository:

git clone https://github.com/TakatoHonda/sui-lang.git
cd sui-lang

REPL Example

>> = v010
>> + v1 v05
>> . v1
15
>> .exit

The example demonstrates variable assignment ( = v010), arithmetic ( + v1 v05), and output ( . v1).

Repository

The project is hosted on GitHub at the following URL (no hyperlink): https://github.com/TakatoHonda/sui-lang. It currently has roughly 400 stars and 16 branches.