The specification behind ESAC, a program of compact benchmarks that measure difficult capabilities without measuring budget. ESAC-GI is released and runnable; ESAC-AG has a harness and one of its seven categories. Items are generators rather than stored questions, and every score carries the version tag of the instrument that produced it.
documentation benchmark constitution evaluation specification mit-license reproducibility benchmark-suite contamination llm esac evals llm-evaluation agentic-ai general-intelligence ai-benchmark llm-benchmark agentic-evaluation ecowestern
-
Updated
Oct 19, 2026 - JavaScript