Epoch AI finds six frontier models cannot yet automate its own research and operations work

Epoch AI reports that six AI models, given 11 real tasks from its own operations, cannot yet fully automate its work.

Tasks included graphic design, data insights, and research design; Epoch ran the models at the highest available reasoning settings without intervention, then graded outputs manually against its employee standards. Fable 5.1 and GPT-6 Astra led on average performance, reliably handling well-defined work such as coding and computational analysis.

All models still failed on open-ended judgment, including meeting Epoch's standards, designing informative experiments, and generating diverse ideas. On that basis, Epoch concludes AI cannot yet replace its workers. The findings are presented as the first Epoch Automation Reports, and Epoch says it will expand the task suite and publish updated results as new models are released.