What specific benchmarks have Grok 3 and GPT-4o been tested on

- Mathematics: Grok 3 achieved a score of 93.3% on the 2025 American Invitational Mathematics Examination (AIME), while GPT-4o's performance in this area is not explicitly highlighted in the available data. However, Grok 3's performance in the AIME 2024 was noted as significantly higher than GPT-4o's general performance in math-related tasks[1][3][5].

- Science and Reasoning: Grok 3 scored 84.6% on the GPQA (Graduate-Level Expert Reasoning) test, showcasing its strength in graduate-level science knowledge. GPT-4o's performance in similar reasoning tasks is generally lower compared to Grok 3[1][3][6].

- Coding: Grok 3 achieved 79.4% on the LiveCodeBench, outperforming GPT-4o in code generation tasks. GPT-4o's specific score on LiveCodeBench is not detailed, but Grok 3 generally excels in coding benchmarks[1][3][5].

- General Knowledge: Grok 3 scored 79.9% on the MMLU-Pro, which tests broad knowledge across multiple subjects. GPT-4o scored 72.6% on the same benchmark, indicating Grok 3's stronger performance in general knowledge tasks[3].

- Multimodal Understanding: Grok 3 demonstrated capabilities in multimodal tasks like MMMU (Multimodal Multitask Model Understanding), though specific comparisons with GPT-4o in this area are limited[1][3].

Overall, Grok 3 tends to outperform GPT-4o in specialized tasks such as mathematics, science, and coding, while GPT-4o might be more versatile in general-purpose applications[2][5][6].