OpenAI has launched LifeVal, a new benchmark designed to measure how effectively AI models help people complete everyday tasks. Announced on November 2, 2026, the evaluation covers practical activities such as finding childcare, managing household finances, preparing meals, shopping, organising caregiving and accessing government services.
Unlike traditional AI benchmarks that focus on mathematics, coding or academic knowledge, LifeVal evaluates whether an AI assistant can understand a person’s needs, navigate tools and websites, and produce a useful real-world outcome. OpenAI says it developed the benchmark using research on daily activities and a survey of 3,995 people to identify where users most want help from AI. The company has also published example results from its latest models, including scores above 90% on several everyday workflows.
Key Takeaways
- OpenAI introduced LifeVal on November 2, 2026.
- The benchmark measures AI performance on practical, everyday responsibilities.
- Its task selection was informed by a survey of 3,995 people.
- Evaluation areas include caregiving, health, education, personal finance, shopping and government services.
- OpenAI published example scores of 90.4% to 100% across several practical workflows.
- One example involving kindergarten selection received a 93.1% score.
- A flood-preparation volunteer coordination task received 98%.
- LifeVal focuses on completing tasks rather than simply generating conversational answers.
- OpenAI says it has used the evaluation to guide improvements over the past year.
What Is OpenAI’s LifeVal Benchmark?
LifeVal is an evaluation framework designed to measure how useful AI systems are in ordinary situations.
Most established AI benchmarks measure specific technical abilities. They may test whether a model can solve mathematical problems, write computer code, answer scientific questions or understand complex language.
However, these abilities do not necessarily demonstrate whether an AI assistant can help someone manage everyday responsibilities.
For example, knowing how to create a household budget is different from actually helping someone organise their expenses, compare options and prepare a usable financial plan.
Similarly, explaining how to find childcare is different from researching available providers, comparing their services and preparing a shortlist based on a family’s requirements.
LifeVal is intended to measure this second type of capability.
OpenAI wants to evaluate whether AI assistants can understand personal requirements, use available tools and complete practical workflows that produce meaningful results.
How Did OpenAI Develop LifeVal?
OpenAI says it used existing research on how people spend their time to identify the activities that matter in everyday life.
The company began with two major datasets:
- American Time Use Survey: A US government-backed dataset that tracks how people spend their time across different activities.
- Multinational Time Use Study: An international collection of time-use surveys covering daily activities across different populations.
OpenAI then expanded this research with a new demographically representative survey involving 3,995 respondents.
Participants were asked where they would most value assistance from a personal AI assistant.
The resulting evaluation includes tasks across several areas:
| Category | Example activities |
|---|---|
| Caregiving | Organising care and supporting family members |
| Health | Preparing meal plans and managing health-related responsibilities |
| Education | Finding suitable schools and preparing for learning |
| Household administration | Organising schedules and managing responsibilities |
| Personal finance | Creating budgets and comparing financial options |
| Shopping | Finding products and organising grocery orders |
| Government services | Navigating public services and paperwork |
This approach makes LifeVal different from benchmarks built primarily around academic questions or standardised technical problems.
Its task selection is based on activities that people actually want help completing.
How Does LifeVal Test AI Models?
LifeVal evaluates practical workflows rather than isolated questions.
A task may require an AI assistant to understand several requirements, research information, navigate websites, use tools and prepare a final result.
For example, a family moving to a new area might ask an AI assistant to compare kindergartens near its new home and shortlist the best options.
The system needs to understand the family’s requirements, research available institutions and organise the information into a useful shortlist.
Another task could involve preparing a community volunteer group ahead of a flood. This requires the assistant to understand the situation, organise responsibilities and produce an actionable plan.
These examples demonstrate the type of work OpenAI wants its benchmark to measure.
The company also says it has used LifeVal over the past year to track improvements in browser use, computer use and connectors.
These capabilities allow ChatGPT to interact with websites, work with information and use connected services rather than relying exclusively on text generation.
OpenAI Publishes High Scores on Everyday Tasks
Alongside the announcement, OpenAI published several examples showing how its latest models performed on LifeVal tasks.
Some of the reported results include:
| Everyday task | Published score |
|---|---|
| Find and shortlist suitable kindergartens after moving | 93.1% |
| Organise community volunteers ahead of a flood | 98.0% |
| Find a suitable new home for a family | 100% |
| Arrange a remote way to watch a game with a grandparent | 96.3% |
| Prepare a meal and care plan for a parent adjusting to a diabetes diagnosis | 90.4% |
<text weight=”medium” color=”default”>Selected LifeVal task scores published by OpenAI</text>
<Chart content={{“chartType”:”bar”,”meta”:{“description”:”Selected task-level results published by OpenAI. These are individual workflow scores, not an overall benchmark average.”},”xKey”:”task”,”series”:[{“dataKey”:”score”,”label”:”Published score”,”valueSuffix”:”%”}],”yAxisMin”:0,”yAxisMax”:100,”layout”:”vertical”,”data”:[{“task”:”Find a family home”,”score”:100},{“task”:”Organise flood volunteers”,”score”:98},{“task”:”Watch a game remotely”,”score”:96.3},{“task”:”Find a kindergarten”,”score”:93.1},{“task”:”Prepare a care meal plan”,”score”:90.4}]}}/>
OpenAI also listed several additional workflows with best recorded GPT-6 Astra scores of 100%, including:
- Finding and booking a suitable sensory class for a toddler.
- Checking vehicle servicing requirements and booking an inspection.
- Ordering groceries within a budget and preparing a spreadsheet.
- Finding a replacement engagement ring using a photograph and diamond report.
- Finding and ordering a compact device to replace several personal gadgets.
- Unsubscribing from email lists.
These figures are examples published by OpenAI. They should not be interpreted as an overall accuracy score across every LifeVal task or as independently verified results.
Why Everyday AI Tasks Are Becoming Important
The launch reflects a major change in how technology companies are developing AI assistants.
Early generations of chatbots were primarily designed to answer questions, explain concepts and generate content.
Newer AI systems are increasingly expected to take actions.
A personal assistant may need to search for a service, compare prices, fill out information, organise documents and prepare a decision for the user.
This requires capabilities beyond language understanding.
AI systems need to interact with websites, understand user preferences, remember relevant context and use external tools.
LifeVal attempts to measure these capabilities through practical tasks.
For OpenAI, this is particularly relevant as ChatGPT expands its browser, computer-use and connected-app capabilities.
The benchmark could also help the company identify weaknesses that traditional coding and reasoning tests might not reveal.
LifeVal Could Influence the Future of AI Assistants
A benchmark can influence how AI models are trained and evaluated.
If developers measure performance on real-world tasks, they can identify where models struggle and improve their ability to complete those workflows.
For example, an AI assistant might be good at researching information but fail to organise it into a usable result. Another model might understand a user’s requirements but struggle to navigate a website.
LifeVal gives OpenAI a framework for measuring these types of problems.
However, there are limitations to consider.
The public announcement does not provide a complete independently reproducible evaluation suite or enough methodological detail to treat every published percentage as a universal measure of real-world reliability.
Performance can also vary depending on the user’s location, available services, quality of information and complexity of individual requests.
The results therefore provide an indication of progress on selected tasks rather than proof that AI can reliably manage every everyday responsibility without human oversight.
The Bigger Picture
OpenAI’s LifeVal launch highlights an important direction for the AI industry: moving from measuring what models know to measuring what they can actually accomplish.
Academic benchmarks remain valuable for testing reasoning, mathematics and technical knowledge. But they do not fully capture the practical challenges involved in helping people manage their lives.
As AI assistants gain access to browsers, computers and connected services, their ability to complete multi-step tasks will become increasingly important.
LifeVal is an attempt to measure that progress using activities that ordinary people recognise and value.
Looking Ahead
The next challenge will be making everyday-task benchmarks more transparent, independently reproducible and representative of different countries and populations. Since many tasks depend on local services, languages and regulations, broader international testing will be important before results can be generalised across users.
For OpenAI, LifeVal could become an important internal measurement tool as ChatGPT develops into a more capable personal assistant. Its long-term significance will depend not just on high benchmark scores, but on whether users experience more reliable help with real responsibilities in their daily lives.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
