ResearchBenchmarksReasoning on ARC Challenge (Norm Log Acc, Compression)Follow63.5Normalized Log AccuracyT-Free57.7859.26560.7562.235Mar 16, 2026Evaluation ResultsMethodMethodLinksNormalized Log AccuracyCompression (Bytes/Position)T-Freeevaluation_protocol=25...evaluation_protocol=25-shot2026.0363.55.51HATifiedevaluation_protocol=25...evaluation_protocol=25-shot2026.0362.15.51Llamaevaluation_protocol=25...evaluation_protocol=25-shot2026.03584.92