Fewer tokens, same meaning.

Compression for everything you send a model: data, prompts and code.