An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run

Epoch AI has launched the new MirrorCode benchmark, which tests AI models' ability to recreate complete programs without access to the original code. Claude Opus 4.7 leads with a 56% solve rate, successfully rebuilding a 16,000-line toolkit in just 14 hours. The MirrorCode benchmark is crucial for assessing AI models' efficiency in complex programming tasks. Despite the impressive performance of Claude Opus 4.7, all tested models failed on the most complex tasks, highlighting ongoing challenges in the field.