A new report from nonprofit METR updates its time horizon plot, revealing that AI capabilities continue growing exponentially, with model-completed tasks doubling every four months. However, the evaluation shows models remain less proficient at open-ended tasks requiring judgment compared to verifia