Gemini Pro excels in common sense reasoning tasks, and research from Stanford and Meta suggests that Gemini is underrated. Researchers designed a new test set and found that Gemini's performance in complex reasoning tasks is significantly stronger than before. The evaluation includes 11 language-based datasets and one multimodal dataset, showing that Gemini Pro is on par with GPT-3.5. While GPT-4Turbo leads in overall performance, Gemini Pro slightly outperforms in specific tasks. Inference experiments demonstrate that Gem