OpenAI published LifeVal on 2 November 2026 to measure how its models perform on everyday tasks in caregiving, health, education, household administration, personal finances, shopping and government services.
OpenAI chose tasks using the American Time Use Survey and the Multinational Time Use Study, which track how people spend their time. It then surveyed 3,995 demographically representative respondents about where they most wanted help from a personal assistant. The evaluation reports model performance as an average reward across tasks.
Finding a kindergarten after a move could involve comparing nearby options and organising enrolment details, rather than stopping at a single answer. That is the work in one of OpenAI’s example tasks.
On OpenAI’s evaluation, GPT-5.2 scored 49.3% average reward and GPT-6 Astra scored 80.6%. The models were released on 11 December 2025 and 3 September 2026 respectively. OpenAI lists a 93.1% score for the kindergarten example, without assigning it to a particular model, and 100% scores for GPT-6 Astra on tasks including booking a sensory class and unsubscribing from email lists.
OpenAI says it has trained its models over the past year on browser use, computer use and connector tasks, capabilities it is also enabling in ChatGPT.
