ResearchBenchmarksCode Generation on APPS-GCode Generation on APPS-GFollow68.01Pass Rate BeforeACW64.609566.3097568.0169.71025Apr 17, 2026Evaluation resultsMethodMethodLinksPass Rate BeforePass Rate AfterDegradation (%)ACW2026.0468.0167.8120MATRIX2026.0468.0167.8714