AI Evaluation Shifts Toward Real-World Task Execution: OpenAI Launches LifeVal to Measure Personal Assistant Capabilities

OpenAI has released the LifeVal benchmark to measure AI performance in everyday scenarios such as caregiving, health, education, household management, personal finance, and government services. Unlike benchmarks focused on knowledge Q&A and isolated capabilities, it focuses on whether AI can advance tasks to an executable, deliverable state and solve real-world problems. The benchmark references the American Time Use Survey and was built based on a survey of 3,995 respondents; OpenAI has continuously used it for tracking over the past year.