ResearchBenchmarksBucket Sort on Formal-language benchmark lengths 41-500 (test)Follow99.4AccuracyPR-LSTM25.0444.34563.6582.955May 16, 2026Evaluation ResultsMethodMethodLinksAccuracyPR-LSTM2026.0599.4LSTM2026.0599.3TransformerArchitecture=encoder-onlyArchitecture=encoder-only2026.0591.9SRNNArchitecture=Stack-RNNArchitecture=Stack-RNN2026.0578.1TRNNArchitecture=Tape-RNNArchitecture=Tape-RNN2026.0570.7RNN2026.0527.9