LifeVal is a benchmark designed to assess how well AI models handle high‑value, personalized consumer tasks that a digital super‑assistant could automate. It spans 148 task categories—such as finance management, moving, job searching, and childcare—and represents over $1.7 trillion in annual willingness‑to‑pay. To build the activity distribution, the authors surveyed more than 3,800 demographically representative consumers and cross‑referenced the results with the American Time Use Survey (2025) and the Multinational Time Use Study (2025). For the most valuable activities, task creators crafted evaluation tasks using real ChatGPT conversations, memory files, and personal context. Creators had to meet an expertise threshold—completing the task repeatedly and using it in real life for over a year—and write a rubric to ensure the evaluation captures the best personalized completion. Every task underwent at least three rounds of human quality control before being admitted to the final evaluation set.\n\nReview