Shorter proofs
Reduce proof tokens while keeping the same theorem statement.
In development · Seeking pilot teams
Shorter Lean proofs.
Every change checked.
Vexorium’s upcoming agent refactors Lean 4 proofs to reduce their length and checking work, while preserving the theorem and checking compatibility across supported versions.
For teams that build and maintain formal proofs in Lean.
Built for proof maintenance
Less checking work
Reduce heartbeats, Lean’s deterministic measure of the work needed to check a proof.
Version compatibility
Check the same proof across supported Lean and library versions, with compiler evidence for each.
Lean Refactor Arena · Public warm-up
#1 on the closed-source LLM track
Entry: Vexorium-Agent · 15 public warm-up problems · Snapshot as of October 20, 2026
- 93.02
- Combined warm-up score
- 88.7%
- Shorter proofs than the long references, on average
- 90.4%
- Less checking work than the references, on average
- 100%
- Of warm-up proofs compile on every listed Lean version
These are warm-up results, not final competition standings. The entry is a development portfolio built from many runs, without the track’s US$3-per-problem budget limit. Final rankings depend on the organizers reproducing submitted code under that budget on the full benchmark.
Research
The research behind the product
Accepted · MATH-AI 2026 workshop at NeurIPS 2026
The Verification Gap
Auditing Proof Progress and Repair on IMO 2026
An AI judge can spot the error. The acceptance rule can still let it pass.
This study examines how grading rules turn model judgments into acceptance decisions. A score that allows repairable errors can admit an incomplete proof. For autonomous research, the rule that authorizes progress deserves its own audit.
Work with us
Bring us a research challenge whose results matter.
IT challenges
You know what needs to work. Getting there is the hard part.
From stubborn integrations to performance bottlenecks, we investigate the constraints and test a practical path forward.
Business challenges
You see the impact on your business, but the problem is hard to pin down and a practical solution remains out of reach.
We help define the problem, investigate its causes, and test possible solutions against the outcomes that matter to you.
Future challenges
You worry about what comes next, even if you cannot yet name the challenge.
We explore plausible scenarios, test assumptions, and identify useful steps you can take now.