Microsoft Research introduced StudentSim on October 12 as a research system. It trains personalized student simulators from publicly available, de-identified learner records in chess, second-language English writing and mathematics. The simulators are designed to model how a learner might answer questions and change a response after receiving tutor guidance. Microsoft stresses that StudentSim is not a product and is not deployed in classrooms.

In the Microsoft Research post, StudentSim scored 0.5150 for behavioral fidelity and 0.9067 for guidance responsiveness in chess; the corresponding scores were 0.2316 and 0.7186 for GPT-5.4, and 0.4535 and 0.2721 for the chess-move model Maia2. In a reinforcement-learning experiment, a tutor trained with StudentSim feedback reached 90.5% accuracy in expert evaluation, versus 75.7% for a comparison without reinforcement learning and 71.6% for a comparison using GPT-5.4 simulation. The researchers say testing with real learners remains necessary to assess learning outcomes, safety and usability.