<ul><li>The study compares OpenAI's o1-mini and o3-mini models on math problems, finding that the o3-mini outperformed the o1-mini with fewer reasoning chains.</li><li>Response accuracy declined as the reasoning chains grew, indicating that 'thinking harder' isn't the same as 'thinking longer'.</li><li>Newer reasoning models use compute more effectively, resulting in a smaller accuracy drop for unsolvable problems.</li><li>OpenAI plans to unify the o-series and GPT-series models with the release of GPT-5, and the o3-mini model will not be released as a standalone model.</li></ul>

OpenAI’s o3-Mini Shows Superior Accuracy Than o1-Mini Without ‘Thinking’ Longer: Harvard Study

Discover more