ResearchBenchmarksMissing Duplicate on Formal-language benchmark lengths 41-500 (test)Follow100AccuracyTRNN50.39263.27176.1589.029May 16, 2026Evaluation ResultsMethodMethodLinksAccuracyTRNNArchitecture=Tape-RNNArchitecture=Tape-RNN2026.05100PR-LSTM2026.05100TransformerArchitecture=encoder-onlyArchitecture=encoder-only2026.0556.4SRNNArchitecture=Stack-RNNArchitecture=Stack-RNN2026.0555.2LSTM2026.0554.3RNN2026.0552.3