ResearchBenchmarksGeneral Reasoning on GPQA D (Accuracy, Loss)Follow27.3AccuracyAdam18.4620.75523.0525.345Apr 10, 2026Evaluation ResultsMethodMethodLinksAccuracyLossAdamModel scale=3BModel scale=3B2026.0427.31.975NexusModel scale=3BModel scale=3B2026.0423.41.957AdamModel scale=1BModel scale=1B2026.0418.82.191NexusModel scale=1BModel scale=1B2026.0418.82.172