Frontier data for AI that works in finance.
Move32 builds RL environments, expert training data, and private evaluations for financial reasoning. Our team spent two decades auditing banks. Now we teach machines to do the same work, and we measure which ones can.
Three products, one discipline.
Everything we ship is graded by code, checked by auditors, and kept away from training data until you pay for it.
RL Data
Expert reasoning traces, injected error corpora, and grading rubrics built from real financial statements by senior auditors. Made for post-training, priced per task.
Talk to us about dataRL Environments
Financial statement review as a verifiable environment. An agent reads a full annual report, hunts for errors we planted, and a deterministic engine scores every finding.
Built on the same stack as our benchPrivate Evals
A held-out test set we run for you. Your model or agent takes the exam on our infrastructure. The answer key never leaves the building, so scores stay honest.
Request an evaluationWhich systems can actually review a financial statement?
Every model and agent below reviewed the same set of statements with expert-injected errors. We grade their findings against a private answer key.
No published results yet. The leaderboard appears once a run is published.
Notes from the lab.
We publish what we learn about evaluation, reward design, and financial reasoning. No press releases, just work.
Working on financial AI? So are we.
Labs use our environments and data to train. Banks and funds use our private evals to decide what to deploy. Either way, the first conversation is free.
[email protected]