Most AI benchmarks test puzzles, exam questions or coding problems. OpenAI's LifeVal points in a more useful direction: it grades AI on the dull admin of daily life, the kind of task people actually want help with. If assistants are ever going to earn their keep, it will be on finding childcare, filling in a government form or sorting a budget, not on contest maths, so a benchmark aimed at real chores is the right instinct.
The way OpenAI chose the tasks is the strongest part. Rather than guess, it drew on two large time-use studies, the American Time Use Survey and the Multinational Time Use Study, which catalogue how people actually spend their days across different groups, times and seasons. It then ran a demographically representative survey of 3,995 people to find where they would most value help from a personal assistant. The resulting tasks span caregiving, health, education, household administration, personal finances, shopping and accessing government services, with agents using browsers, computers and connectors to carry them out.
The trouble is what OpenAI does not show. The post displays a handful of example task scores between 90.4% and 100%, but it names no models, gives no total task count, and does not describe how answers are graded. We review ChatGPT, the product this is meant to improve, and those high numbers are easy to read as a claim that it can already run your errands. They are not evidence of that. A score only means something when you know which model earned it, on how many tasks, and by what measure, and here all three are missing.
There is also no outside validation. LifeVal is an internal evaluation OpenAI says it has used over the past year to drive its own models, which makes it a tuning target as much as a test. A benchmark the same company builds, scores and reports has an obvious incentive to flatter, and nothing here lets a reader check the numbers independently.
Still, the direction deserves credit. Measuring whether an assistant can navigate the paperwork and decisions of ordinary life is more honest about what these tools are sold for than another knowledge quiz. For now LifeVal is best read as a statement of intent and a hint at where OpenAI is steering ChatGPT, not as proof that the job is done. The number worth waiting for is an independent one, on named models, with the task set and grading method in the open.