OpenAI has published LifeVal, a new benchmark measuring how well AI carries out concrete everyday-life tasks such as childcare, moving, medical procedures, and household budgeting, designed using the U.S. Bureau of Labor Statistics' "American Time Use Survey," the international "Multinational Time Use Study," and a survey of about 4,000 people. Published model scores are 49.3% for GPT-5.2, 73.0% for GPT-5.4, 80.0% for GPT-5.6, and 80.6% for GPT-6 Astra, with gains slowing between generations: GPT-5.2 to GPT-5.4 improved by 23.7 points, while GPT-5.6 to GPT-6 Astra rose by only 0.6 points. Test items include choosing a kindergarten after a move, running a community volunteer organization in preparation for floods, and finding housing that fits a budget and set of conditions. LifeVal emphasizes "agentic execution" involving multi-step operations, including referencing a user's past conversation history and personal information, operating a browser to narrow candidates, and presenting matching options. OpenAI prepared a separate evaluation model to verify whether operations were completed and introduced a mechanism disqualifying attempts to exploit back-end code as shortcuts, with multiple reviews by human experts built in. OpenAI stated that for artificial general intelligence (AGI) to benefit all of humanity, it is necessary to measure not only academic evaluations but also