o3-mini Achieves Higher Accuracy Than o1-mini Without Longer Reasoning Chains

A systematic analysis of reasoning chain length across o1-mini and o3-mini variants on the Omni-MATH benchmark finds that o3-mini (m) achieves superior accuracy without requiring longer reasoning chains than o1-mini. Accuracy generally declines as reasoning chains grow across all models and compute settings, even when controlling for question difficulty. The accuracy drop is significantly smaller in more proficient models, suggesting newer generations of reasoning models use test-time compute more effectively. While o3-mini (h) achieves a marginal accuracy gain over o3-mini (m), it does so by allocating substantially more reasoning tokens across all problems, including those o3-mini (m) can already solve. The findings provide new insights into the relationship between model capability and reasoning length, with implications for efficiency, scaling, and evaluation methodologies.