By Paul Derrington, Jr.
CEO | Chief Engineer | Research Engineer | Editor, D-CSIL
3 min read
Key Takeaways
OpenAI published LifeVal on November 2, an evaluation of everyday tasks such as finding childcare, accessing government services, and ordering groceries. On OpenAI’s chart, GPT-6 Astra scored 80.6% average reward, up from 49.3% for GPT-5.2 in December 2025. OpenAI built and scored the evaluation, charts only its own models, and has not published the task set.
What LifeVal measures
OpenAI published LifeVal on November 2, 2026. OpenAI describes it as an evaluation of everyday tasks such as finding childcare, accessing government services, creating a budget, and ordering groceries. OpenAI writes: “To be a truly helpful personal assistant, ChatGPT should be able to take actions that actually get things done.”
In one example OpenAI shows, the task is to find a home in Luzerne County that fits a budget, has a separate living unit, and keeps the commute short after school drop-off — and to add the deed to the county records cart so the user can review it. In OpenAI’s example response, the model reports that it found one matching home and added the deed to the cart for review.
How OpenAI picked the tasks
OpenAI started from two studies of how people actually spend their time: the American Time Use Survey and the Multinational Time Use Study. It added a new survey of 3,995 people that OpenAI calls demographically representative, asking where they most wanted a personal assistant’s help. The resulting tasks cover seven areas: caregiving, health, education, household administration, personal finances, shopping, and accessing government services.
OpenAI says it used LifeVal over the past year to track and drive improvement in model performance on everyday tasks, and that it trained its models on browser use, computer use, and connector tasks. OpenAI says more than one billion people use ChatGPT every week, and that it wants to move ChatGPT “beyond a chat tool and towards a truly agentic personal assistant.”
The reported scores
OpenAI’s chart plots one number per model, an unweighted average reward across LifeVal tasks, labeled “Average reward (unweighted).” GPT-5.2 scored 49.3% in December 2025. GPT-5.4 reached 73.0% in March 2026, GPT-5.5 reached 77.4% in April, GPT-5.6 hit 80.0% in July, and GPT-6 Astra sits at 80.6%. GPT-5.4 was 23.7 points above GPT-5.2; the three later releases added 7.6 points together, and Astra is 0.6 points above GPT-5.6.
OpenAI also shows five example tasks with a score each, describing them as how “our current models” perform, without naming a model per task: enrolling in the right kindergarten after a move, 93.1%; managing volunteers ahead of a flood, 98.0%; finding a new home, 100%; setting up a way to watch a game with grandma from far away, 96.3%; meal prep for a parent adjusting to a new diabetes diagnosis, 90.4%. OpenAI also lists six tasks where GPT-6 Astra’s best recorded attempt hit 100%, including booking a car inspection before a road trip and ordering halal groceries within a budget. A best recorded score is the best attempt, not the average.
SemiAnalysis published its separate GDPx benchmark of expert tasks the same day, with a top score of 12%. GDPx is a third-party benchmark of expert work and LifeVal is OpenAI’s own evaluation of everyday tasks, so the scores measure different things.
What OpenAI has not published
OpenAI built the evaluation, chose the tasks, ran the scoring, and charts its own models only; no Claude, Gemini, Grok, or Muse model appears. OpenAI’s page does not publish the task set, the number of tasks, or how “average reward” is graded, and there is no public leaderboard.
AI Weekly noted that because OpenAI has not released the task set or opened scoring to other models, every number on the chart is self-graded. OpenAI has not said whether it will publish the task set.