PutnamBench
A benchmark dataset for evaluating the formal mathematical reasoning capabilities of AI, based on problems from the Putnam mathematics competition.
Timeline (last 90 days)
Lean 4 and its accompanying mathematics library Mathlib have become the de facto standard for formalized mathematics, and benchmarks such as arXivLean and PutnamBench are built on top of them.
On benchmarks shared with published prover systems, Maestro solved 669/672 problems on PutnamBench, 59/60 on Lean-IMO-Bench, and achieved 6/6 on IMO 2026.
PutnamBench contains problems from the William Lowell Putnam Mathematical Competition and is a classic benchmark for measuring automated theorem proving capability.
The Lean community has built a large mathematics library, mathlib, and datasets such as MiniF2F and PutnamBench have emerged specifically to test AI's formal reasoning capabilities.
All Facts (4)
Lean 4 and its accompanying mathematics library Mathlib have become the de facto standard for formalized mathematics, and benchmarks such as arXivLean and PutnamBench are built on top of them.
50%UnverifiedOn benchmarks shared with published prover systems, Maestro solved 669/672 problems on PutnamBench, 59/60 on Lean-IMO-Bench, and achieved 6/6 on IMO 2026.
50%UnverifiedPutnamBench contains problems from the William Lowell Putnam Mathematical Competition and is a classic benchmark for measuring automated theorem proving capability.
50%UnverifiedThe Lean community has built a large mathematics library, mathlib, and datasets such as MiniF2F and PutnamBench have emerged specifically to test AI's formal reasoning capabilities.
50%