EasyProver solves near-saturated Lean benchmarks and half of harder Aegis tasks
Original title: Scaling Lean Theorem Proving for Real-World Code Verification
On a stratified Aegis sample of 2,000 statements, it proves 1,054 within a 30-minute budget.
EasyProver is a Lean proving system of recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service. With an unnamed off-the-shelf frontier model and no extra training, it proves 669 of 672 PutnamBench problems, all six IMO 2026 formalizations, 597 of 600 NTP4VC conditions, and all 189 VERINA tasks. Aegis converts competitive-programming problems into harder checks over real Rust code translated to Lean with Aeneas.
Unconfirmed: The frontier model used by EasyProver is not named.
- Aegis produces 2,045 validated Rust-Lean pairs and 11,533 candidate specifications.
- Obligations cover fixed-width arithmetic, explicit failure, generated loops, casts, and mutation.
- EasyProver proves 1,054 of 2,000 sampled statements, or 52.7 percent.
- The authors also propose the tasks as reinforcement-learning rewards.
From the original article
1 min read at huggingface.co
We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service.
By Debargha Ganguly, Divya Raghunathan, Dejan Jovanović · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.