OpenAI has a new way to grade its AI: not on exams, code puzzles or maths contests, but on the chores of everyday life. The company published LifeVal on November 2, a benchmark designed to measure how well an AI helps people find childcare, balance a budget, order groceries and get through government paperwork — the practical work that fills most people's days.
The headline number is flattering for OpenAI itself. Its newest model, GPT-6 Astra, scored 80.6% average reward across LifeVal tasks, up from 49.3% for GPT-5.2 in December 2025. GPT-5.6, the previous generation, sat at 80.0%. The catch is one OpenAI does not hide: it built the test, chose the tasks and ran the scoring, and it publishes scores for its own models only.
Built from how people actually spend their time
LifeVal's task list comes from two large time-use studies — the American Time Use Survey and the Multinational Time Use Study — plus a new survey of 3,995 people that OpenAI describes as demographically representative, asking where they most wanted a personal assistant's help. The resulting areas cover caregiving, health, education, household administration, personal finances, shopping and government services.
The tasks ask the AI to finish the job, not just give advice. Examples include enrolling a child in the right kindergarten, organising flood volunteers, finding a new home, planning how to watch a game with grandma and preparing meals for diabetes. OpenAI reports its current models scoring 93.1% on the kindergarten task, 98.0% on flood volunteers, 100% on finding a new home, 96.3% on watching the game with grandma and 90.4% on diabetes meal prep. It also says the best recorded GPT-6 Astra runs hit 100% on some tasks, including booking a sensory class, a car-service inspection, halal groceries and unsubscribing from email lists — best runs, not averages.
Why a chores benchmark matters
LifeVal extends a shift already visible in OpenAI's earlier GDPVal benchmark, which graded AI on economically valuable office work. The message is that the next contest between AI labs is not who tops an academic leaderboard, but whose assistant actually gets things done in daily life — browsing, using computers and acting through connectors on the user's behalf. OpenAI says it has used LifeVal internally for a year to track and drive improvement in its models on exactly these tasks.
That framing is also why the benchmark's limits matter. OpenAI's announcement gives no task count, no plain-language breakdown of how tasks are graded and no independent verification of the numbers, and an average reward score is not the same as a share of tasks finished correctly in the real world. Everyday tasks can have broad requirements and subjective outcomes, so a strong benchmark score should not be read as independent proof of real-world reliability.
The fine print under the headline score
For now, LifeVal is best read as a statement of intent as much as a measurement: OpenAI wants its models judged on whether they help people navigate decisions, manage responsibilities and accomplish meaningful tasks, not just answer questions. The 80.6% figure shows how far its own models have come on its own test in under a year — from 49.3% to 80.6% between GPT-5.2 and GPT-6 Astra. Whether rival labs adopt the benchmark, or publish their own scores on it, will decide whether LifeVal becomes a shared yardstick or remains a self-graded report card.
The AI race is moving out of the exam hall and into the kitchen, the inbox and the council office — and the winner will be the assistant that finishes the errand.
