On this page
OpenAI published LifeVal on its blog with a page date of November 2, 2026, and posted about it on X at 2026-11-02 17:27:26 UTC, which is 2026-11-03 01:27:26 Asia/Shanghai. The post comes from @OpenAI. LifeVal is framed as a new benchmark for everyday personal-assistant work—tasks people want help finishing, not only chatting about.
The X note says the benchmark was shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant, naming examples such as finding childcare, accessing government services, and managing personal finances. Separately, the OpenAI blog describes a demographically representative survey of 3,995 respondents used to learn where people most want assistant help. Those are two different inputs: the X post’s user-feedback framing, and the blog’s 3,995-person survey. Before that survey, the blog says the team also looked at the American Time Use Survey (ATUS) and the Multinational Time Use Study (MTUS) as catalogs of everyday activity.
Related on-site context: Codex Usage Shows gpt-6.2-astra Before Any Official Listing, Tibo Says Dots Will Expand to GPT-6 Astra and GPT-6.1 Terra, Artificial Analysis Intelligence Index Moves to v5 — Private Benchmarks Now Dominate, and ChatGPT Adds iCloud Mail and Calendar Plugins.
What LifeVal targets
The blog’s subtitle is Measuring AI progress on the tasks that matter in everyday life. Mission framing on the page ties AGI benefit to practical daily challenges: finding childcare, accessing government services, creating a budget, preparing for school, modifying recipes, ordering groceries, designing workout plans, and more. The stated bar for a personal assistant is the ability to take actions that get things done, not only advice or a chat reply.
Task coverage listed on the blog spans caregiving, health, education, household administration, personal finances, shopping, and accessing government services.
How OpenAI says it has used the eval
OpenAI writes that it built the evaluation to measure model performance over time on everyday tasks, and that it has used this evaluation to track and drive improvement over the past year. On the capability side, the blog says models were trained to become faster, more capable, and more aligned on browser use, computer use, and connectors, helping move ChatGPT from a chat tool toward an agentic personal assistant.
The X post matches that timeline language: over the past year, LifeVal was used to measure and improve how well models handle these everyday tasks. The post links back to the LifeVal blog.
Task examples recorded on the blog
The LifeVal page shows individual workflow examples with scores. Figures below are as recorded on the OpenAI blog’s task examples—we haven't re-run them here:
| Example (blog wording, shortened) | Blog-recorded score |
|---|---|
| Enroll in the right kindergarten after moving | 93.1% |
| Manage volunteers ahead of a flood | 98.0% |
| Find a new home for my family | 100% |
| Watch a game with grandma from far away | 96.3% |
| Meal prep for a parent adjusting to a new diabetes diagnosis | 90.4% |
A separate table on the same page is headed Best recorded GPT‑6 Astra score. Several rows there show 100%, including sensory-class booking, car inspection booking, budgeted halal grocery orders into a spreadsheet, engagement-ring replacement from a photo and diamond report, a pocket device order, and email unsubscribes. Those Astra figures are blog table entries. They are not DigIntoAI lab results, not a third-party leaderboard verification, and not a claim that a full public task suite can be downloaded and re-run end to end from this announcement alone.
LifeVal itself is an evaluation and product narrative from OpenAI—not a model catalog update. Nearby Astra naming stays on the linked on-site pieces.
Separating LifeVal from nearby claims
- Not a new model ID drop. LifeVal is a benchmark and progress narrative, not a replacement for a model listing or shipping notice.
- Not a pricing-table update. Neither the blog nor the X post publishes a new plan matrix or price sheet.
- Not a fully open, independently runnable suite (from what is shown). The blog presents example workflows and recorded scores. We haven't seen a complete open task set with instructions for outsiders to run the full suite themselves.
- Astra “best recorded” scores ≠ our testing. When those percentages appear here, they are labeled as OpenAI blog recordings.
Product pages, help docs, and social posts can land at different times; the facts above follow the original blog and X post.
Quick facts
- Who. OpenAI, via the LifeVal blog and @OpenAI on X.
- When. Blog page date November 2, 2026; X post 2026-11-03 01:27:26 Asia/Shanghai (2026-11-02 17:27:26 UTC).
- What. LifeVal — benchmark for everyday personal-assistant tasks.
- Inputs (two sources). X: feedback from thousands of ChatGPT users. Blog: ATUS / MTUS background, then a demographically representative survey of 3,995 respondents.
- Domains. Caregiving, health, education, household administration, personal finances, shopping, government services.
- Past year. Used to track and drive improvement; training emphasis on browser use, computer use, and connectors.
- Scores. Blog example table and Best recorded GPT‑6 Astra score rows — OpenAI’s recordings; we haven't tested them here.
- Open suite. Full open task set for independent end-to-end runs: not seen in the materials used for this write-up.
In short
OpenAI’s 2026-11-03 early-morning Asia/Shanghai X post, paired with the November 2, 2026 LifeVal blog, introduces a benchmark aimed at everyday personal-assistant work—shaped by thousands of ChatGPT users (X) and a 3,995-person survey (blog), with ATUS/MTUS as earlier background. OpenAI says the eval has already guided a year of measurement and improvement, especially around browser use, computer use, and connectors as ChatGPT moves toward acting, not only chatting. Blog task percentages and Astra “best recorded” cells are OpenAI’s published examples—not DigIntoAI re-tests, and not a claim that the full suite is publicly runnable.