We believe that how we teach mathematics should be transparent, reproducible, and improvable by researchers and educators everywhere. QLM is committed to contributing to the digital public goods that advance math education for every student.
The model and evaluation framework are open. The measurement engine and training data that make it all work stay proprietary — ensuring continuous improvement for every student who uses it.
An open-source math tutoring model. It asks questions rather than giving answers. It detects errors in student reasoning and responds with targeted Socratic questions.
The model is trained to ask questions, not give answers. It detects student errors and responds with targeted Socratic questions. Optimized for learning, not helpfulness.
Most math tutoring models are evaluated on answer correctness or conversation fluency. Neither measures whether the student actually learned. We evaluate against published academic benchmarks — and publish the results, including where we fall short.
A stronger proprietary measurement engine produces a stronger open model. We open-source the model and evaluation framework. The measurement engine and training data that make it all work stay proprietary — ensuring continuous improvement for every student who uses the system.
Researchers can reproduce our results, challenge our methodology, and build on our work. Claims about math tutoring efficacy should be verifiable, not vendor promises.
Students furthest from opportunity often attend schools that cannot afford proprietary tutoring software. The open model ensures that the foundation of effective math tutoring is accessible to everyone.
Researchers studying misconception detection, productive struggle, and Socratic tutoring can contribute improvements that benefit every student using the system.
Districts evaluating math tutoring software deserve to see how the model works and how it is evaluated. Open source is the strongest possible answer to those questions.
What stays proprietary: The measurement engine (patent pending) that trains the model and evaluates whether tutoring works. The engine is what makes the open model better over time.
Honest disclosure: We publish benchmark results transparently — including where the model falls short. Evaluation results are at quantumlearningmachines.com/research/external-benchmark-results.
We are actively seeking district partners for classroom pilots to evaluate the tutor with real students. If you are a researcher, district, or organization interested in collaborating: research@quantumlearningmachines.com