ResearchBenchmarksLanguage Modeling on Public Pretraining Dataset (train)Follow1.33LossAdam1.329681.331841.3341.33616Apr 10, 2026Evaluation ResultsMethodMethodLinksLossAdamModel scale=3BModel scale=3B2026.041.33AdamModel scale=1BModel scale=1B2026.041.331NexusModel scale=1BModel scale=1B2026.041.338NexusModel scale=3BModel scale=3B2026.041.338