OpenAI introduces LifeVal, a benchmark for AI help with everyday tasks

OpenAI has published LifeVal, a benchmark for measuring how well AI models help people with practical daily tasks. Examples include finding childcare, creating a budget and handling government services. The company says the tasks were chosen based on what people most want help with. It used the benchmark to track model performance on everyday tasks over the past year.

The task list started from two existing time-use studies: the American Time Use Survey and the Multinational Time Use Study. Those datasets record what people do across demographics, times of day and seasons. OpenAI then ran a new survey of 3,995 respondents asking where everyday people most wanted help from a personal assistant. The resulting tasks cover caregiving, health, education, household administration, personal finances, shopping and accessing government services.

The page includes examples with a percentage score beside each one. A task to compare kindergartens near a new home and shortlist the best fits scored 93.1%. A request to find a home in Luzerne County that fits a budget and keeps a commute short scored 100%. Other examples include organizing flood preparation volunteers (98.0%), arranging a remote game-watching setup with a grandmother (96.3%) and building a meal plan for a parent with a new diabetes diagnosis (90.4%). The page does not say how these scores are calculated or how many attempts each one reflects, so the figures are best read as OpenAI's own sample rather than a full results table.

The page frames the tasks as actions rather than answers. It says ChatGPT should "take actions that actually get things done," and the example responses show the model proposing a plan, adding a deed to a cart for review, or asking the user to confirm a choice before proceeding. OpenAI ties this to recent training work on browser use, computer use and connectors. The company says that work helps move ChatGPT "beyond a chat tool" toward an agentic personal assistant.

OpenAI presents the benchmark as a way to measure progress on tasks that matter to ordinary people, not only on academic tests of knowledge. The page's own wording is that the company is "hopeful" LifeVal is a step in that direction. The page does not describe how the tasks are graded beyond the percentages, and it does not name the models behind the scores in the examples.