The numerical accuracy used to store a model's weights and calculations—higher precision (like float32) is more accurate but uses more memory, while lower precision (like int4) is more efficient but less precise.
Multi-step reasoning, logic puzzles, mathematical problem-solving