OpenAI's LifeVal Benchmark Tests AI on Everyday Errands, With Only OpenAI Models Scored
OpenAI says GPT-6 Astra averages 80.6% on its everyday-tasks benchmark, up from 49.3% for GPT-5.2, but has not released the tasks or the grading method.
Published
Last updated / Fact-checked on
OpenAI on November 2 published LifeVal, a benchmark built to measure how well its models complete everyday errands such as finding childcare, booking a car inspection and ordering groceries on a budget. OpenAI's chart puts GPT-6 Astra at an 80.6% average reward, up from 49.3% for GPT-5.2, according to coverage of the release. The task set and grading method have not been released.
Photo: Pixabay
OpenAI says it built LifeVal because, "to ensure AGI benefits all of humanity, we need to measure and improve model progress on what matters in the lives of everyday people." It said it has used the evaluation "to track and drive improvement in model performance on everyday tasks over the past year."
LifeVal at a glance
| Item | Detail |
|---|---|
| Published | November 2, 2026, on OpenAI's site |
| Task sources | American Time Use Survey, Multinational Time Use Study, new 3,995-person survey |
| Areas | Caregiving, health, education, household admin, personal finance, shopping, government services |
| Headline score | GPT-6 Astra 80.6% vs. GPT-5.2 49.3% (unweighted average reward, per reports of OpenAI's chart) |
| Models scored | OpenAI models only |
| Released for others | No task set, grader or rubric published |
How OpenAI chose the tasks
OpenAI started with two large studies of how people spend their time: the American Time Use Survey and the Multinational Time Use Study. It then ran a new, "demographically representative" survey of 3,995 people. The survey asked where they would most want help from a personal assistant.
The tasks are meant to end in action, not advice. "To be a truly helpful personal assistant, ChatGPT should be able to take actions that actually get things done," OpenAI wrote. It noted that more than one billion people use ChatGPT each week.
The examples OpenAI shows
The page lists five worked examples with scores for "our current models," without naming a specific model:
- Enrolling in the right kindergarten after a move: 93.1%
- Organizing volunteers ahead of a flood: 98.0%
- Finding a family home that meets budget and commute limits: 100%
- Setting up a way to watch a game remotely with a grandparent: 96.3%
- Planning meals for a parent with a new diabetes diagnosis: 90.4%
A second table lists six tasks with a "best recorded GPT-6 Astra score" of 100%. They include booking a toddler's sensory class, booking a car inspection, ordering halal groceries on a budget, finding a replacement engagement ring and unsubscribing from email lists. "Best recorded" means these are top results, not typical ones.
The headline numbers
CellCog and Digg reported that OpenAI's chart shows an unweighted average reward of 80.6% for GPT-6 Astra and 49.3% for GPT-5.2. CellCog also listed scores read from the chart for models in between: GPT-5.4 at 73.0%, GPT-5.5 at 77.4% and GPT-5.6 at 80.0%. DailyMech could not view the chart directly, so these figures are attributed to those reports.
If those figures are right, most of the gain came early. GPT-6 Astra sits only 0.6 points above GPT-5.6.
What OpenAI leaves out
LifeVal is OpenAI's own test, run on OpenAI's models. The page does not say how many tasks it contains, define "average reward," explain the grading, or say how many attempts sit behind each score. No models from other labs appear. OpenAI also says it used LifeVal to drive improvement for a year, so its models were tuned toward this benchmark as well as measured by it.
That makes LifeVal a useful look at what OpenAI is aiming for. It is a weak basis for comparing ChatGPT with rival assistants. For more on how OpenAI sets out what its models should do, see our report on its Model Spec update.
What to watch
- Release. Whether OpenAI publishes the task set or grader so others can run LifeVal.
- Rival scores. Whether other labs or independent testers report results.
- Live-site testing. Whether tasks run on real websites or test copies.
More coverage is on our Artificial Intelligence hub.
How we reported this
DailyMech read OpenAI's LifeVal page directly for its methodology, examples and quotes. The overall average-reward figures come from OpenAI's chart as reported by CellCog and Digg, because the chart did not load in our copy of the page. All scores are OpenAI's own and have not been independently reproduced.
Published and fact-checked November 2, 2026.
Sources
- OpenAI, "LifeVal," Nov. 2, 2026: https://openai.com/index/lifeval/
- CellCog, "OpenAI LifeVal: GPT-6 Astra Hits 80.6% on Everyday Tasks," Nov. 2, 2026: https://cellcog.ai/blog/openai-lifeval/
- Digg, "OpenAI introduces LifeVal to measure AI performance on everyday tasks," Nov. 2, 2026: https://digg.com/ai/08jg9bxp
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.