ByteDance's Seed team recently announced the impressive results of the 2025 Gaokao comprehensive subject test: the Douyin Dabao Seed 1.6-Thinking model scored 683 in liberal arts and 648 in science, meeting the admission line for Tsinghua and Peking University, and performing outstandingly in Gaokao tests among AI models both domestically and internationally.
The test used the national new volume I and Shandong provincial independent examination papers. Dabao competed against five top AI models, including Google's Gemini 2.5 Pro, DeepSeek R1, and OpenAI o3. Dabao achieved the highest score of 683 in liberal arts, while its science score of 648 was second to Google's Gemini 2.5 Pro with 655. In individual subjects, Dabao achieved the highest scores in six subjects: Chinese, English, Physics, History, Geography, and Politics, and also scored over 140 in mathematics, demonstrating excellent logical reasoning ability.
In this AI "Gaokao battle," different models showcased their strengths in various subjects. DeepSeek R1 achieved the highest math score of 145, Gemini 2.5 Pro obtained the highest chemistry score of 82, and OpenAI o3 tied with Gemini 2.5 Pro for the highest biology score of 77. This differentiated performance reflects the unique characteristics of different AI models in knowledge structure and reasoning methods.
The Seed team discovered an important technical detail: during the first test, due to the low clarity of the online exam paper, all models lost many points in subjects that relied on images such as chemistry and biology. After obtaining high-resolution test pictures, the team retested using a combined text and image approach. The results showed that Dabao improved by nearly 30 points in chemistry and biology, resulting in a total science score of up to 676. This finding verified the importance of full-modal reasoning in unlocking model potential and provided important insights for the development of AI in visual understanding and cross-modal reasoning.
Shandong uses the "3+3" Gaokao system, where Chinese, Math, and English are basic subjects, and the other three selected subjects use a grade-based scoring system. According to analysis by experienced high school teachers in the region, Dabao's grade-based score combination could reach around 690, ranking approximately in the top 80 according to the 2025 Shandong one-point distribution table, which is sufficient to compete for top universities like Tsinghua and Peking University. This performance not only demonstrates Dabao's strong capabilities but also highlights its adaptability in complex scoring systems.
Dabao's outstanding performance in the Gaokao not only showcases its extensive knowledge base and reasoning abilities, but more importantly, it verifies the great potential of multimodal AI in complex cognitive tasks. Especially in handling science questions that combine text and images, Dabao demonstrated a nearly 30-point improvement, providing a new technical path for AI applications in education. This achievement marks that domestic large AI models have reached international advanced levels in comprehensive cognitive abilities, laying a solid foundation for deep application of AI in the field of education.
