OpenAI has released a framework for measuring the long-term effects of AI tools on student learning, developed with the University of Tartu and the SCALE Initiative at the Stanford Accelerator for Learning.
The Learning Outcomes Measurement Suite was created in response to research showing that existing studies capture performance on individual exams but do not reveal how AI changes the way students learn over time.
OpenAI ran a randomised trial with more than 300 college students preparing for neuroscience and microeconomics exams, assigning participants to either a no-AI control group using standard online resources or to one of two variants of its study mode tool.
Results were mixed by subject: students using study mode scored roughly 15% higher than the control group in microeconomics, while neuroscience showed a positive but statistically inconclusive difference.
The suite combines learning interaction classifiers, quality graders, longitudinal outcome measures and standardised cognitive assessments to track model behaviour, learner responses and cognitive development across extended periods.
Related reading
- OpenAI extends single-minus amplitudes to gravitons
- OpenAI publishes GPT-5.3 Instant system card
- OpenAI strikes classified AI deal with US military and calls for industry-wide terms
OpenAI said all data is de-identified and that the system can incorporate external data such as exam scores or attendance records supplied by partner institutions.
The framework is being validated through large-scale randomised trials, including a study in Estonia involving nearly 20,000 students aged 16 to 18 conducted over several months.
The ChatGPT owner said it intends to publish findings from those trials and release the suite as a public resource.
The recap
OpenAI launches framework to measure AI’s impact on learning.
Nearly 20,000 students aged 16-18 study the suite in Estonia.
Validation underway through randomized trials and planned public release.