DeepSeek-Prover-V2

model
DeepSeek
RL for subgoal decomposition in formal mathematical reasoning. Includes the DeepSeek-ProverBench evaluation suite.
Paper (arXiv)GitHubHuggingFaceDeepSeek-ProverBench

Model Details

Architecture MOE
Parameters 671B

Paper

arXiv: 2504.21801

reasoningopen-weight

Related