ResearchBenchmarksGeneral Reasoning on GPQA (Accuracy, Loss)Follow25.8AccuracyAdam24.96825.18425.425.616Apr 10, 2026Evaluation ResultsMethodMethodLinksAccuracyLossAdamModel scale=1BModel scale=1B2026.0425.82.28NexusModel scale=3BModel scale=3B2026.0425.82.047NexusModel scale=1BModel scale=1B2026.04252.261AdamModel scale=3BModel scale=3B2026.04252.062