List your company

Scientific Reasoning on Principia OOD (test)

83.78Average Score @8CoT Self-Instruct
31.67645.20358.7372.257Jun 24, 2026

Evaluation results

MethodLinks
CoT Self-InstructCategory=ARB, Items=47, Training Data Source=CoT Self-Instruct2026.0683.7891.491.87
Combined (2× data)Category=ARB, Items=47, Training Data Source=Combined2026.0683.2493.621.33
Qwen3.5-4BCategory=ARB, Items=47, Training Data Source=Base2026.0681.9191.49—
Agentic Self-InstructCategory=ARB, Items=47, Training Data Source=Agentic Self-Instruct2026.0680.3291.49-1.59
Agentic Self-InstructCategory=Physics, Items=110, Training Data Source=Agentic Self-Instruct2026.0667.0579.090.83
Combined (2× data)Category=Physics, Items=110, Training Data Source=Combined2026.0666.2572.730.03
Qwen3.5-4BCategory=Physics, Items=110, Training Data Source=Base2026.0666.2276.36—
CoT Self-InstructCategory=Physics, Items=110, Training Data Source=CoT Self-Instruct2026.0665.9177.27-0.31
Agentic Self-InstructCategory=SuperGPQA, Items=1324, Training Data Source=Agentic Self-Instruct2026.0656.8264.950.82
Combined (2× data)Category=SuperGPQA, Items=1324, Training Data Source=Combined2026.0656.565.110.5
CoT Self-InstructCategory=SuperGPQA, Items=1324, Training Data Source=CoT Self-Instruct2026.0656.2865.330.28
Qwen3.5-4BCategory=SuperGPQA, Items=1324, Training Data Source=Base2026.065665.41—
Agentic Self-InstructCategory=Overall, Items=2113, Training Data Source=Agentic Self-Instruct2026.0651.4759.911.04
Combined (2× data)Category=Overall, Items=2113, Training Data Source=Combined2026.0651.1759.910.74
CoT Self-InstructCategory=Overall, Items=2113, Training Data Source=CoT Self-Instruct2026.0651.160.060.67
Qwen3.5-4BCategory=Overall, Items=2113, Training Data Source=Base2026.0650.4359.54—
Agentic Self-InstructCategory=RealMath, Items=632, Training Data Source=Agentic Self-Instruct2026.0635.4343.671.75
CoT Self-InstructCategory=RealMath, Items=632, Training Data Source=CoT Self-Instruct2026.0635.2543.671.57
Combined (2× data)Category=RealMath, Items=632, Training Data Source=Combined2026.0634.9744.31.29
Qwen3.5-4BCategory=RealMath, Items=632, Training Data Source=Base2026.0633.6841.93—