PutnamBench — 670 of 672
The same approach as IMO 2026, across all 672 formal statements of PutnamBench: 99.7% verified, and first place on the leaderboard we do not control.
670 / 672Formal statements verified end to endResult
#1On the official leaderboard, confirmed by the PutnamBench teamRank
3Independent gates a proof had to pass to be countedMethod
What counts as solved
A problem counted only if the Lean file passed every gate — the Lean kernel, Comparator and AXLE, all three agreeing. Any one of them alone is a weaker claim, and a proof that satisfies one checker and not another is exactly the sort of thing a long agentic run produces if nobody is looking.
The two that are not closed are not closed. They are drawn as such everywhere we show this.
Check it yourself
The official leaderboard is maintained by the PutnamBench team. Re-verification needs the AXLE API, which wants Python 3 and a network connection and nothing else.