OpenAI describes LifeVal benchmark for everyday AI tasks

OpenAI says LifeVal is intended to assess whether models help people navigate decisions, manage everyday responsibilities and get meaningful things done.[1] The company contrasts this aim with academic assessments focused on model knowledge, saying LifeVal goes beyond those evaluations.[1] OpenAI also says a helpful personal assistant should take actions that get things done.[1] That framing makes the benchmark about completing practical tasks, rather than only answering questions; OpenAI presents it as one way to measure progress on work that matters in everyday life.[1]

OpenAI says LifeVal draws on the American Time Use Survey and the Multinational Time Use Study, supplemented by a demographically representative survey of 3,995 respondents.[1] It says the survey asked where people most wanted help from a personal assistant.[1] The resulting task areas include caregiving, health, education, household administration, personal finances, shopping and government services.[1] Examples include finding childcare, creating a budget, preparing for school, modifying recipes, ordering groceries and planning workouts.[1] This gives the benchmark a broad scope across personal and administrative needs.

OpenAI’s examples show tasks that involve organizing information or preparing next steps. In a kindergarten search, the system shortlists three options and organizes enrollment details, while leaving two preferences for the user to decide.[1] A flood-preparation example produces an action plan for volunteers to assess sites and prepare for severe weather.[1] A home-search example finds a property that meets the user’s requirements and calculates the transfer tax.[1] In a diabetes-care example, the output includes meals, reminders and questions for a care team, with review requested before sharing.[1]

OpenAI says it has used LifeVal over the past year to track and drive model improvement on everyday tasks.[1] It also says it trained models to become faster, more capable and more aligned on browser use, computer use and connector tasks.[1] OpenAI says enabling those capabilities in ChatGPT expands the tasks it can help with and moves it toward an agentic personal assistant.[1] Separately, AI Affairs reports average reward scores of 49.3% for GPT-5.2 and 80.6% for GPT-6 Astra.[2] Aivio reports that OpenAI has not identified which model produced five example scores or published the dataset or scoring rubric; it also reports that LifeVal has no published leaderboard or independent run.[3]