Epoch tests AI with real-world work tasks: top models scored an average of 65%, meaning they are not yet capable of automating their daily jobs.

Epoch AI presents its Epoch Automation Reports, which delegate daily tasks such as graphic design and research project planning to AI, judging them by the same criteria applied to human staff. Top-tier models like Claude Fable 5.1 and GPT-6 Astra averaged a score of 65%. This means they met only 65% of the scoring items considered standard for humans, which still counts as task failure.