OpenAI's own test scores its models on everyday jobs like finding a kindergarten
OpenAI has published LifeVal, a test of how well its models finish tasks such as ordering groceries or planning meals after a diabetes diagnosis. The scores are OpenAI's own, for its own models.
One task in OpenAI's new LifeVal test asks an AI model to compare kindergartens near a family's new home. In the reply OpenAI shows, the model narrows the choice to three and leaves two preferences for the user to decide. OpenAI gives that task a score of 93.1%.
OpenAI published LifeVal on 2 November 2026. A benchmark is a standard set of tasks used to score AI systems, and OpenAI describes LifeVal as a way of measuring progress on everyday-life tasks. It says it built the test itself, around the areas where users wanted the most help from AI.
To choose the tasks, OpenAI says it started from two large studies of how people spend their time: the American Time Use Survey and the Multinational Time Use Study. It added a new survey of 3,995 people, which it describes as demographically representative. The tasks cover caregiving, health, education, household administration, personal finances, shopping and accessing government services.
A second example involves a parent adjusting to a new diabetes diagnosis. OpenAI shows the model producing a first-week plan with meals, reminders and questions for the care team, scored at 90.4%. OpenAI also lists tasks where its GPT-6 Astra model's best recorded score was 100%, including ordering halal groceries for three favourite meals within a budget and itemising the order in a spreadsheet.
OpenAI says it used LifeVal both to track and to drive improvement in its models on everyday tasks over the past year. The company says a helpful assistant must take actions that get things done, not just give advice, and that it trained its models to become much faster and more capable on tasks where a model uses a web browser or a computer.
OpenAI says the aim is to go beyond academic tests of what models know and to measure how well AI models can help people with decisions and everyday responsibilities. On its LifeVal page it writes: "We are hopeful that LifeVal represents a step towards that goal."
OpenAI's chart shows its models' average score on everyday-life tasks across five releases, from 11 December 2025 to 3 September 2026. The figures are 49.3% for GPT-5.2, 73.0% for GPT-5.4, 77.4% for GPT-5.5, 80.0% for GPT-5.6 and 80.6% for GPT-6 Astra. These are OpenAI's own figures for its own models.
Sources
- Lifeval · openai.com