AI & agents
OpenAI releases LifeVal, an eval for everyday agentic tasks
- Published
- Source
- OpenAI blog
Summary
OpenAI has released LifeVal, a new evaluation that measures how well its models complete practical everyday tasks as a personal assistant rather than just answering questions. The task list draws on the American Time Use Survey, the Multinational Time Use Study, and a new demographically representative survey of 3,995 respondents, covering caregiving, household administration, personal finances, shopping, and access to government services. On example tasks such as comparing kindergartens or planning meals after a diabetes diagnosis, OpenAI says its latest models scored between 90.4% and 100% (figures from OpenAI's own evaluation).
Why it matters for our work
The eval shifts measurement from what models know to what they actually get done. It offers a yardstick for how far personal-assistant agents have come on real-life tasks.
Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.