ResearchBenchmarksCode Generation on HumanEval (Score (%), CP len, ∆ CP (%))Code Generation on HumanEval (Score (%), CP len, ∆ CP (%))Follow93Functional Score (%)MaAS92.074492.314792.55592.7953Jan 15, 2026Evaluation resultsMethodMethodLinksFunctional Score (%)CP LengthDelta Completion (%)MaAS2026.01931,810.8—LAMaS2026.0192.111,042.7-42.4