Summary
OpenAI says LifeVal evaluates AI models on practical everyday workflows, with task priorities informed by time-use datasets and a survey of 3,995 respondents. The company reports high GPT-6 Astra scores on selected tasks.
OpenAI has described LifeVal, an evaluation it built to track how well AI models handle practical tasks from everyday life. The company says it used the evaluation over the past year to track and drive improvements, with the aim of measuring more than a model’s ability to answer questions: whether it can help people manage responsibilities and get tasks done.
How LifeVal’s tasks were selected
OpenAI says it began with the American Time Use Survey and the Multinational Time Use Study, which catalogue how people spend time in the United States and internationally. It then added a new demographically representative survey of 3,995 respondents, asking which tasks people would most value help with from a personal assistant.
The resulting task areas include caregiving, health, education, household administration, personal finances, shopping and access to government services. Examples on the LifeVal page range from finding a suitable class for a toddler and arranging a car inspection to ordering groceries within a budget. These are multi-step workflows involving practical requirements, rather than isolated questions.
What the reported scores show
OpenAI’s page lists GPT-6 Astra scores of 100% for several tasks, including finding and booking a sensory class that fits a toddler’s needs and family schedule, arranging a car inspection before a road trip, and ordering groceries within a budget. The page also gives examples of model responses to other scenarios, such as preparing a shortlist of kindergartens and organising volunteers ahead of a flood.
OpenAI links recent model improvements to browser use, computer use and connectors—capabilities that can let an AI system work with websites, computer interfaces and connected services. That direction matters because an assistant that can carry out a sequence of actions may help with tasks that a text-only answer cannot complete on its own.
LifeVal is designed around tasks people said they wanted help with, giving it a practical focus distinct from evaluations centred on academic knowledge. The scores are OpenAI’s reported results for its selected tasks. The page does not explain how the percentages are calculated, so they should be read as scores within this evaluation rather than as a direct estimate of success across everyday use.