与GPT-4O相比，Grok 3在数学推理中表现出了出色的表现。在2025年美国邀请赛数学考试(AIME)中，Grok 3的得分为93.3％，表现明显优于GPT-4O，在类似的数学推理任务中得分79％[1] [2]。这表明Grok 3在复杂的数学问题解决方案中表现出色，这要归功于其先进的推理能力以及使用强化学习随着时间的推移来完善解决方案的能力[1] [5]。相反，尽管GPT-4O在细微的问题解决和上下文理解方面很强，但它并未像Grok 3 [2]那样明确地显示其推理过程。总体而言，Grok 3的Think Mode提供了透明的逐步思考过程，使其对STEM专业人员和教育工作者特别有益[2] [3]。
引用：
[1] https://x.ai/blog/grok-3
[2] https://writesonic.com/blog/grok-3-vs-chatgpt
[3] https://writesonic.com/blog/what-is-grok-3
[4] https://codingmall.com/ knowledge-base/25-global/256724-how-does-grok-3s-performance-compare-compare-compare-to-gpt-4o and-gemini
[5] https://timesofindia.indiatimes.com/technology/technology/tech-news/elon-musks-musks-xai-announces-grok-3-ink-think-think-th--th-and-rok-3-mini-thin-in--in-in--in-in--in-in--in--in-in-in-in-in---------noustial-models/articleshow/118420916一下.cms
[6] https://www.leanware.co/insights/grok-3-vs-gpt-models-comparison
[7] https://latenode.com/blog/grok-3-unveiled-features-capabilities-and-future-future-future-future-future-of-Xais-flagship-model
[8] https://writesonic.com/blog/grok-3-review