ResearchBenchmarksCode Correction on HumanEval+Code Correction on HumanEval+Follow24.2Pass@1CDLM17.9619.5821.222.82Dec 17, 2025Evaluation resultsMethodMethodLinksPass@1CDLMRefinement steps=4, Co...Refinement steps=4, Confidence threshold=0.92025.1224.2BaseRefinement steps=4, Co...Refinement steps=4, Confidence threshold=0.92025.1219.6MDLMRefinement steps=4, Co...Refinement steps=4, Confidence threshold=0.92025.1218.2