GCFDrop-in JSON replacement for all AI pipelines, with superpowers for graph-shaped data.

The most token-efficient wire format for LLMs. 79% fewer input tokens than JSON. 63% fewer output tokens. GCF averages 90.7% comprehension accuracy across 10 models and 3 providers where JSON averages 53.6% and TOON averages 68.5%. Spec v2.0 Stable.

Get Started

Try the Playground

GCF vs TOON

Proven lossless JSON encoding

decode(encode(value)) == value for every JSON value. Verified across 200M+ random round-trips and 7.9M fuzz executions. Six implementations, cross-language conformance matrix verified. Spec v2.0 Stable.

79% fewer input tokens

JSON uses 53,341 tokens at 500 symbols.
GCF uses 11,090. Positional fields, local IDs, and hierarchical grouping eliminate per-record overhead.

63% fewer output tokens

LLMs produce valid GCF with a 3-line primer. 33% smaller output than TOON. 5/5 generation validity on every frontier model across 3 providers. Zero training.

The format LLMs understand without training

90.7% comprehension, 5/5 generation validity across every frontier model from Anthropic, OpenAI, and Google. No model has seen GCF in training. TOON's own decoder rejects LLM-generated output on 7 of 9 models. 23 comprehension runs, 28 generation runs, zero losses.

141 conformance fixtures, 6 languages

All implementations at v1.0.0+. 141 conformance fixtures. 200M+ property-based round-trips across all languages. Cross-language 5x5 encode/decode matrix verified. CLIs for all 6 languages. Spec v2.0 designated Stable.

Beats TOON on TOON's benchmark

Wins all 6 datasets on TOON own benchmark. 34% fewer tokens on mixed-structure, 42% on semi-uniform, 3% on flat. Their datasets, their tokenizer, their library.