GCF: A Token-Optimized Wire Format for Structured LLM Interactions

GCF is an AI-native wire format for structured data in LLM systems. 2,400+ evaluations across 11 models and 3 providers. 43 billion+ lossless round-trips across 5 formats. Two published papers.

Comprehension: 100% accuracy on every frontier model on standard workloads. 91.2% on structurally complex data where JSON drops to 54.1% and TOON to 68.8%. No model has ever been trained on GCF.

Token efficiency: 50-92% fewer tokens than JSON (50-69% per call, up to 92% with session deduplication). JSON overhead is 81% structural noise at scale; GCF overhead is 0.2%.

Tokenizer safety: GCF's grammar characters were selected from the set empirically verified to have near-zero merge rates across 43 tokenizers from 20 providers. Pipe delimiter: 0.47% merge rate, 24 mergeable words. JSON: 8.17%, 1,939 mergeable words. TOON tab: 32.91%, 1,238 mergeable words. A controlled experiment proves delimiter choice matters at the transformer level: models trained with merge barriers on delimiter characters develop 4.6x more structural attention heads, achieve 3x better structured data comprehension and 3-5x better code comprehension, with zero natural language cost. See companion paper: "Merge Barriers in BPE Tokenization" (DOI: 10.5281/zenodo.20925910).

Six implementations (Go, TypeScript, Python, Rust, Swift, Kotlin), 157 conformance fixtures, zero dependencies. Specification v3.1 Stable. Lossless across 17 serialization formats.