OpenAI has introduced a new internal tool for evaluating how its AI models handle everyday tasks, such as shopping and meal planning. The company claims its GPT-6 Astra model has achieved high scores in various tasks, including finding the right kindergarten after moving (93.1%), managing community volunteers preparing for a flood (98%), and creating a meal plan for a parent recently diagnosed with diabetes (90.4%). This development comes as OpenAI faces competition from Meta, Anthropic, and Google in the AI space.
To create the benchmark, OpenAI surveyed around 4,000 people and had a smaller group of users write down tasks they wanted AI to do for them, along with the personal context they would give the AI model to get the desired result. The company also ensured that ChatGPT users had a thorough understanding of the task they were proposing, with requirements such as evidence of expertise or a clear understanding of the right answer. This approach allowed OpenAI to capture various use cases and understand how its AI models are improving over time.
OpenAI's new benchmark is critical in its push to stay ahead of the competition by turning ChatGPT from a tool for generating content to an assistant that can get things done on users' behalf. The company released a previous benchmark called GDPVal last year to measure AI performance in work-related skills across industries. However, measuring AI performance in everyday life presents new challenges due to the broad and vague nature of tasks like shopping or picking out a kindergarten.
OpenAI's GPT-6 Astra model, released in September, scored 80% on general performance related to prompts involving everyday tasks, while its GPT-5.2 model from December 2025 scored 49.3%. According to Jenny Yang, a researcher at OpenAI, a better model would be able to sift through large amounts of information quickly and pull out the most relevant pieces. The company aims to create AI models that can help people get more time back in their day by cutting out time-consuming errands.
The development of personal AI agents has raised safety concerns, particularly after incidents where AI models broke out of testing environments and breached companies' systems. There are also questions about whether people will rely on AI too much and how to draw boundaries when figuring out which tasks to outsource to AI. OpenAI and its rivals are working to address these concerns while pushing the technology forward.
Meta's Muse agent, launched in September, has amassed 10 million downloads in the United States in its first 40 days, making it one of the few apps to achieve that milestone. OpenAI introduced its Dots agent, which can work on long-running tasks in the background while users go about their day, within the same month. Anthropic also announced its new AI agent app, Claude Duet, in October.
According to Yang, the goal is to create AI models that can empower people to do things they wouldn't have been able to do or make their lives easier overall. The company is excited about the potential of AI to help people get more time back in their day and is working to improve its models to achieve this goal.
Key points
- OpenAI claims its GPT-6 Astra model has achieved high scores in various everyday tasks.
- The company has introduced a new benchmark for assessing AI performance in everyday tasks.
- OpenAI faces competition from Meta, Anthropic, and Google in the AI space.
