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

GCF is a bidirectional text-based wire format for LLM interactions. 1,300+ evaluations across 10 models from Anthropic, OpenAI, and Google. Comprehension: 90.5% average accuracy (four models at 100%) where JSON averages 53.6% and TOON averages 68.5%. Generation: 5/5 validity on every frontier model, zero training. TOON's official decoder rejects LLM-generated output on 7 of 9 models. Token efficiency: 79% fewer input tokens than JSON, wins all 6 datasets on TOON's own benchmark. Session deduplication (92.7% by 5th call) and delta encoding (81.2%) compound savings. Implemented in Go, TypeScript, Python, Rust, Swift, Kotlin.