The company says task selection draws on two major time-use studies and a representative survey of 3,995 people. The aim is to measure useful actions beyond a conversational answer.
OpenAI says it has used the evaluation for a year to improve its models. The publication describes an internal measurement tool; it does not by itself announce a newly available service.
The results presented are produced by OpenAI. They alone do not establish an assistant’s reliability across real personal situations.