Goedel-Prover-V2 models achieve new open-source SOTA on MiniF2F (up to 90.4% pass@32) and PutnamBench (86 problems solved) via scaffolded synthesis, self-correction, and averaging.
E RL T RAINING DETAILS We further explain our RL training in detail
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction
Goedel-Prover-V2 models achieve new open-source SOTA on MiniF2F (up to 90.4% pass@32) and PutnamBench (86 problems solved) via scaffolded synthesis, self-correction, and averaging.