AI tutors hold the promise of adapting to each student's unique strengths, weaknesses, and preferred guidance. However, collecting real-world evidence about which teaching strategies work for which student is slow, costly, and inherently sparse. Student simulators offer a proxy for this signal, but existing approaches fall short: state-tracking models fit student behavior but struggle to process explanations or corrections, while LLM role-play follows guidance fluently but fails to accurately mirror the student's competence.
To address this gap, researchers from Hugging Face and Microsoft present StudentSim, a training framework that transforms sparse per-student data into individualized simulators. The method employs a two-stage pipeline: first, pooled training across students, followed by per-student specialization. This enables simulators to both replicate a student's own responses and update them appropriately under tutor guidance.
The team also introduces StudentSimEval, a standardized evaluation protocol spanning 60 students across chess, second-language English writing, and mathematics. Using public learner datasets with de-identified records, StudentSimEval measures two key metrics: behavioral fidelity (F), how well a simulator matches a student's responses, and guidance responsiveness (R), how readily it updates under tutor guidance. All methods are fit and evaluated on the same records for fair comparison.
Across all three domains, StudentSim outperforms GPT-5.4 on both metrics. In chess, StudentSim achieves F=0.51 and R=0.91, compared to GPT-5.4's 0.23 and 0.72, and Maia2's 0.45 and 0.27. As a proof of concept, using StudentSim as a reward model for tutor reinforcement learning produces a chess tutor that expert humans rate as more accurate, better-guided, and more personalized than both a no-RL baseline and a tutor trained against a GPT-5.4 simulator reward.
The code is available on GitHub at microsoft/StudentSim, allowing researchers to build on this work and extend the benchmark.