OpenAI published LifeVal on November 2 to track AI progress on practical tasks outside conventional academic tests. Its task selection draws on American and multinational time-use datasets, plus a survey of 3,995 respondents about where an assistant would help.

Examples cover household administration, shopping, caregiving and access to public services. The emphasis is on completing workflows, rather than producing advice alone.

Read the results in context

This is an OpenAI-created evaluation. Its announcement illustrates selected task results but does not establish an independent cross-model leaderboard or an overall score for a named model. Those examples should not be read as guaranteed reliability in a person's real circumstances. We have not independently reproduced the evaluation.