OpenAI introduced LifeVal, an evaluation built to measure how well ChatGPT models perform on practical, everyday tasks that people actually want help with. The motivation is tied to OpenAI‘s stated mission of ensuring artificial general intelligence benefits all of humanity, especially the more than one billion people who use ChatGPT weekly. The article argues that being a helpful personal assistant requires more than conversational advice; ChatGPT needs to take actions that get things done.

To identify which tasks matter most, OpenAI combined two existing time-use studies, the American Time Use Survey and the Multinational Time Use Study, with a new demographically representative survey of 3,995 respondents. That survey asked people where they would most value assistance from a personal assistant. LifeVal tasks cover caregiving, health, education, household administration, personal finances, shopping, and accessing government services. The evaluation is intended to track and drive model improvement on these everyday tasks over time.

The article includes several example tasks with measured model performance. On enrolling a child in the right kindergarten after moving, the model scored 93.1%. Managing volunteers ahead of a flood scored 98.0%. Finding a new home that fits budget and commute constraints scored 100%. Setting up a way for a grandmother and grandchild to watch a game together remotely scored 96.3%. Meal prep for a parent adjusting to a new diabetes diagnosis scored 90.4%. Each example shows the model producing an organized shortlist, action plan, or practical first-week plan rather than just general advice.

OpenAI attributes these results to a year of training focused on making models faster, more capable, and more aligned on browser use, computer use, and connectors tasks. Expanding these capabilities in ChatGPT moves it beyond a chat tool toward an agentic personal assistant. The article closes with the position that AI progress should be measured not only through academic knowledge assessments but also by how well models help people navigate decisions and manage real responsibilities. LifeVal is presented as a step in that direction.