LifeVal

LifeVal is an evaluation published by OpenAI on November 2, 2026 to measure how well AI models and personal agents handle everyday-life tasks: finding childcare, accessing government services, creating a budget, preparing a child for school, modifying recipes, ordering groceries, designing workout plans and more.

Where GDPval focuses on professional work, LifeVal targets household and personal administration. OpenAI says it has used the evaluation over the past year to track and drive progress: GPT-6 Astra averages 80.6%, up from 49.3% for GPT-5.2 (December 2025), and reaches a best recorded score of 100% on 10 of 16 tasks.

How the tasks were chosen

• Starting from the American Time Use Survey (ATUS) and the Multinational Time Use Study (MTUS), which catalog everyday tasks across demographics, times of day and seasons;
• plus a new demographically representative survey of 3,995 respondents about where people most want help from a personal assistant;
• the resulting tasks span caregiving, health, education, household administration, personal finances, shopping and government services.

How it is evaluated

Models must actually get things done like a real assistant, using browser use, computer use and connectors, not just give advice.

Example tasks (from OpenAI's page)

Task	Score
Enroll in the right kindergarten after moving	93.1%
Manage volunteers ahead of a flood	98.0%
Find a home that fits budget and commute, add the deed to the cart	100%
Meal prep for a parent with a new diabetes diagnosis	90.4%
Book a toddler sensory class, book a car inspection, order halal groceries on budget	100% (best recorded)

Limitations

• Only OpenAI's own models are reported; no Claude, Gemini or other labs.
• The task set, rubric and reproduction materials are not public, and "average reward" is not defined.
• Outlets such as CNN note that everyday tasks have broad requirements and subjective outcomes, so benchmark scores are not proof of real-world reliability.

Who it's useful for

• Personal-agent and AI-assistant builders: LifeVal's task categories (childcare, government services, finances, shopping) work as a scenario checklist and an internal-eval reference.
• PMs and investors: a signal of where OpenAI is taking ChatGPT as a personal assistant next.
• Everyday users: a guide to which life chores can already be delegated to ChatGPT.

How to use it

1. Read openai.com/index/lifeval for task categories and examples.
2. Borrow its method (time-use studies plus a user survey) to build a life-tasks eval set for your own agent.
3. Read it alongside GDPval (work tasks) to judge a model across both work and life.
4. Watch whether OpenAI releases the task set and whether third parties report scores for other models.