OpenAI has introduced LifeVal, a benchmark that measures how well its AI models help people complete everyday tasks such as finding childcare, navigating government services, building a household budget and planning meals. Rather than grading chat replies alone, the benchmark tests whether an assistant can take actions that get things done, such as shortlisting kindergartens, preparing enrolment details or adding a deed to a shopping cart for the user to review.
To choose the tasks, OpenAI started with the American Time Use Survey and the Multinational Time Use Study, which catalogue how people spend their days. It then ran a new survey of 3,995 demographically representative respondents to identify where people most want help from a personal assistant. The resulting tasks cover caregiving, health, education, household administration, personal finances, shopping and government services. OpenAI published example scores ranging from 90.4% for a diabetes care and meal plan to 100% for a home-search task, and said its latest models score highly on many LifeVal tasks. The company linked the work to recent gains in browser use, computer use and connectors, which it says are moving ChatGPT beyond a chat tool towards an agentic personal assistant. The scores are OpenAI's own reported results, and the publication does not describe independent testing.
For Australian businesses and professionals, LifeVal changes nothing immediately. It is a research benchmark, not a new product, price or regulation, and it does not alter obligations under the Privacy Act 1988. It is worth noting if staff use AI agents that can access accounts, forms or payments. In that case, avoid entering other people's personal information, such as client, employee or family details, unless your privacy policy allows it, and confirm every submission or purchase before it is final. For low-risk admin, such as drafting a budget or a list of questions for a government service, an assistant may save time, but check the steps against the official source, such as myGov, before acting on them.
Key Points
OpenAI released LifeVal, a research benchmark for measuring AI performance on everyday tasks such as childcare, government services and budgeting.
The task list draws on US and multinational time-use studies plus a new survey of 3,995 respondents.
The benchmark rewards completing actions, not just giving advice, which reflects OpenAI's push toward agentic assistants.
OpenAI reports high scores on many tasks, with published examples from 90.4% to 100%, but these are self-reported.
The work follows recent gains in browser use, computer use and connectors inside ChatGPT.