Webpage · 2026

MirrorCode

Epoch AI

Abstract

Measures: Very long-horizon reimplementation of existing software from behaviour alone. Metric: % hidden tests passed.

BibTeX Copy

@misc{epoch_ai_mirrorcode_2026,
	title = {{MirrorCode}},
	url = {https://epoch.ai/benchmarks/mirrorcode},
	abstract = {Measures: Very long-horizon reimplementation of existing software from behaviour alone. Metric: \% hidden tests passed.},
	author = {{Epoch AI}},
	year = {2026},
	keywords = {agent-eval, gap-sweep-2026-11},
}