We buy courses and exams to create the next generation of evals.
Frontier models have never seen the graded exams that help students structure their thinking. We buy cutting-edge knowledge from the academics who own it, clear the rights, and turn it into the held-out test set labs measure against. The professors get funded for more research.
Université Paris-Saclay
14.2M tokens · 47 evals
Still training on scraped pages? We have knowledge eval sets you can trust.
The reasoning data labs need isn’t hidden. It’s locked: behind university intranets, behind author rights no engineer can sign away.
The exam, not just the lecture
We buy lectures, slides, and video transcripts, but the graded exam and its answer key are the prize. A lecture teaches; the marked exam tests. Everything comes out as clean markdown, organised by subject and difficulty.
Paid for, not scraped
We buy the economic rights from the academics who own them. They keep their moral rights, their name on the work, and can renew.
A clean chain of title
Every token carries documented provenance, from lecture to token. In the EU, where training on unlicensed material is now a live liability, that’s the difference between data a lab can use and data it can’t.
If you’re training a reasoning model and running out of data, we should talk.
We work directly with labs and research teams.
