OpenAI launches LifeVal benchmark and routes Dots assistants to smaller GPT-6.1 Terra

LifeVal is the personal-life counterpart to OpenAI's GDPVal workplace benchmark. 'Because so many people are using the app, we want to capture all the different use cases,' OpenAI researcher Jenny Yang told CNN. The benchmark covers errands such as shopping, budgeting, filling in forms and planning meals. These are the tasks OpenAI wants ChatGPT to handle as it moves from generating content to acting on users' behalf.

The cost side matters just as much. Running personal agents at scale is expensive, so Dots assistants now route work automatically to GPT-6.1 Terra, a smaller model. Fortune asks directly whether compute will limit the push toward personal agents. Terra is also available on Amazon Bedrock at low token prices.

Competitively, Meta, Anthropic and Google are all chasing the same agent market, and Apple's Siri AI is opening to apps this week. Benchmarks like LifeVal let OpenAI define what counts as success.

Caveats: OpenAI designed the benchmark, and scores above 90% on its own tasks invite skepticism. A separate MIT result on GPT-6 Astra interpreting model weights drew pushback: Simon Willison reproduced it with Qwen 3.8 27B on a laptop.