CNN reported that OpenAI announced LifeVal on November 2, 2026, an internal evaluation of AI performance on everyday tasks. The benchmark draws on scenarios such as shopping, finding a kindergarten and planning care. Its scores are results from OpenAI’s own evaluation, as reported by CNN.
What LifeVal evaluates
LifeVal focuses on personal tasks rather than the work-related skills assessed by OpenAI’s earlier GDPVal benchmark, according to CNN. The evaluation includes skills such as navigating websites, using information from a person’s chat history and tailoring conclusions to that person’s needs.
CNN reported three GPT-6 Astra task scores: 93.1% for finding a suitable kindergarten after a move, 98% for coordinating community volunteers preparing for a flood, and 90.4% for creating a meal plan for a parent recently diagnosed with diabetes. These are specific results within OpenAI’s evaluation, not measures of performance across every real-world task.
How OpenAI developed the tasks
CNN reported that OpenAI considered time-use surveys, including the American Time Use Survey and the Multinational Time Use Survey, when selecting everyday activities to assess. The company surveyed around 4,000 people, then asked a smaller group of about 100 to propose tasks, provide relevant personal context and create criteria for judging success.
Task proposers had to show expertise or a strong understanding of what a successful result would look like. The personal context matters: a useful answer to a childcare search or care-planning task depends on the needs and circumstances included in the prompt.
Scores reported for five models
Compute Loading’s video description lists five LifeVal results as unweighted average reward at xhigh effort. In the same reported progression, GPT-5.2 scored 49.3% and GPT-6 Astra 80.6%.
| Model | Reported unweighted average reward |
| GPT-5.2 | 49.3% |
| GPT-5.4 | 73.0% |
| GPT-5.5 | 77.4% |
| GPT-5.6 | 80.0% |
| GPT-6 Astra | 80.6% |
CNN separately reported GPT-6 Astra at 80% for general everyday-task performance and GPT-5.2 at 49.3%. The outlet’s 80% figure and the video description’s 80.6% average-reward figure are reported with different labels.
For more background on the model itself, see NeoTeo’s earlier coverage of GPT-6 Astra.
In its November 4 description, Compute Loading says LifeVal’s task set and grading rubric were not published.