StepFun-Prover-Preview

model
StepFun
Specialized 32B model family for formal theorem proving (Lean 4), using tool-integrated reinforcement learning.
Paper (arXiv)GitHubHuggingFace

Model Details

Architecture DENSE
Parameters 32B
reasoning

Notes

arXiv submission Jul 27, 2025. 70% pass@1 on miniF2F-test.