OpenAI’s LifeVal puts GPT-6 Astra at 80.6% on everyday life tasks
Major
Primary
On November 2, 2026, OpenAI published LifeVal, an evaluation of everyday personal-assistant tasks grounded in time-use surveys and a demographically representative survey of 3,995 respondents. On OpenAI’s unweighted average-reward chart, GPT-6 Astra reaches 80.6%, up from GPT-5.2 at 49.3% on the same series.
At a glance
Figures are from OpenAI’s November 2, 2026 LifeVal publication.
GPT-6 Astra
80.6% average reward (unweighted, OpenAI)
GPT-5.6
80.0% average reward (OpenAI)
GPT-5.2
49.3% average reward (OpenAI)
Survey size
3,995 respondents (OpenAI)
API cost note
GPT-5.6 similar overall to GPT-6 at roughly half estimated API cost (OpenAI)
Speed note
GPT-6 about 23% faster than GPT-5.6 on these tasks (OpenAI)
How to read this claim
Measurement — LifeVal: GPT-6 Astra at 80.6% average reward on OpenAI’s everyday-task chart — OpenAI’s figure; not GDPx and not Index v5.
Bill — API cost note: GPT-5.6 similar overall to GPT-6 at roughly half estimated API cost (OpenAI)
Door — The exact number of LifeVal tasks and the full grading rubric are not listed as a downloadable public suite on the page.
OpenAI published LifeVal with a multi-model everyday-task series that puts GPT-6 Astra at 80.6% average reward.
What it introduces
LifeVal, OpenAI’s everyday life / personal-assistant evaluation.
Who benefits Anyone quoting “80% on everyday tasks” into a product plan, and anyone mixing LifeVal with GDPx or coding benchmarks.
Try this
When a slide says Astra is at 80.6%, ask whether that is LifeVal’s unweighted average reward and whether the comparison is OpenAI’s own chart. Keep cost and latency claims (half cost vs GPT-6; ~23% faster) labeled as OpenAI’s LifeVal API estimates.
What we do not know
The exact number of LifeVal tasks and the full grading rubric are not listed as a downloadable public suite on the page.
Average reward methodology details beyond the published chart are not fully specified in the extractable page text.
Whether third parties can reproduce LifeVal scores is not stated as an open benchmark release.
Evidence
PrimaryLifeVal publication. Scores and survey size are from OpenAI’s November 2, 2026 LifeVal publication.
Independent LifeVal scores are OpenAI’s. The post does not publish a full open task set for independent reruns.
OpenAI published LifeVal on November 2, 2026. The post says the task set was shaped by the American Time Use Survey, the Multinational Time Use Study, and a new survey of 3,995 respondents about where people want assistant help. On the unweighted average-reward chart in the post, GPT-5.2 scores 49.3%, then GPT-5.4 73.0%, GPT-5.5 77.4%, GPT-5.6 80.0%, and GPT-6 Astra 80.6%. OpenAI also says that when LifeVal tasks run in the API, GPT-5.6 reaches similar overall performance to GPT-6 at roughly half the estimated API cost, and that GPT-6 is about 23% faster than GPT-5.6 on these tasks. Example workflows in the post show high Astra scores on individual scenarios such as kindergarten enrollment (93.1%) and flood-volunteer planning (98.0%).
Why it matters
A high score on LifeVal is OpenAI measuring OpenAI on everyday agent work. It is useful for tracking their assistant direction, and it is not a substitute for an independent suite like SemiAnalysis GDPx or Artificial Analysis Index v5.
OpenAI frames LifeVal as measuring progress toward agentic personal assistance beyond chat answers.
Individual task percentages in the examples are OpenAI’s recorded best GPT-6 Astra scores on those workflows.