ResearchBenchmarksLanguage Modeling on Public Pretraining Dataset OODFollow1.606LossNexus1.595721.665111.73451.80389Apr 10, 2026Evaluation ResultsMethodMethodLinksLossNexusModel scale=3BModel scale=3B2026.041.606AdamModel scale=3BModel scale=3B2026.041.623NexusModel scale=1BModel scale=1B2026.041.835AdamModel scale=1BModel scale=1B2026.041.863