Nov 3, 2026
OpenAI's LifeVal tests AI on everyday life admin, graded by OpenAI
LifeVal is OpenAI's own new eval for everyday tasks like childcare, government services and shopping. GPT-6 Astra scores 80.6%, but the task list and grading aren't public.
Sir Toastimir
Images · OpenAI LifeVal page
"Find a home that fits my budget… add the deed to my cart." That's a real example task in LifeVal, OpenAI's own new evaluation.
What it tests
Everyday chores: childcare, government services, budgets, shopping and health. OpenAI says it built the tasks from US and multinational time-use surveys plus a 3,995-person survey.
OpenAI's numbers
Average reward: GPT-5.2 49.3%, GPT-5.6 80.0%, GPT-6 Astra 80.6%. The newest model added just 0.6 points.
Caveats
Not published: the task list, the task count, how grading works, or any other lab's models. So for now, OpenAI grades OpenAI.
Would you let AI put a deed in your cart?
Links
Get your daily slice of AI
One short email with the best new AI tools, robots and AI & tech stock news. Free. Unsubscribe anytime.
We only use your email to send the newsletter.