OpenAI Introduces LifeVal, a Benchmark for Everyday Life Tasks Built From Time-Use Studies and a 3,995-Person Survey

Core Background and Industry Pain Points

Most popular model evaluations test knowledge, math and coding, and say little about what ordinary users care about most: can AI actually get my everyday chores done? OpenAI says more than one billion people use ChatGPT every week, and much of what they want is practical help, such as finding childcare, accessing government services, budgeting, adapting recipes, ordering groceries and planning workouts. On November 2 OpenAI published LifeVal to measure model performance on those tasks, stressing that a helpful personal assistant must take actions that get things done, not just give advice.

Architecture Highlights and Underlying Mechanism

The LifeVal task list was not chosen ad hoc. OpenAI started from the American Time Use Survey and the Multinational Time Use Study, which catalog what people actually do across demographics, times of day and seasons, then ran a new demographically representative survey of 3,995 respondents asking where they most wanted a personal assistant's help. The resulting tasks span caregiving, health, education, household administration, personal finances, shopping and government services. Many require browser use, computer use and connectors, for example finding a home and adding its deed to a county records cart for the user to review. OpenAI says it has used LifeVal for the past year to track and drive model improvements.

Authoritative Benchmarks and Measured Results

All figures are vendor-reported: kindergarten enrollment after a move 93.1%, flood-preparation volunteer coordination 98.0%, a home search within budget and commute limits 100%, remote game-watching setup with grandma 96.3%, and a meal and care plan for a parent with a new diabetes diagnosis 90.4%. OpenAI's table lists GPT-6 Astra best recorded scores of 100% on tasks including booking a toddler sensory class, booking a car inspection, ordering halal groceries within budget with an itemized spreadsheet, finding a replacement engagement ring from a photo and diamond report, and unsubscribing from email lists. The official text we read gives no full task count, overall score or comparison with other vendors' models, so overall figures circulating on X are not used here, and no third-party reproduction exists yet.

Developer Playbook and Getting Started

For teams building agents, LifeVal is mostly useful as a method. First, define tasks from time-use data and user surveys rather than picking whatever is easy to grade. Second, score whether the task was completed, not whether the answer sounds good. You can apply the same idea to an internal eval: list the 20 to 50 real workflows your users run most, write clear acceptance criteria, and require human confirmation before key steps such as payments or form submissions, as OpenAI's example does by leaving the deed in a cart for review. Note that LifeVal appears to be built and scored by OpenAI itself, and the post does not say whether the dataset is public, so be careful with cross-vendor comparisons.