SenseTime Unveils 'Daily New' Fusion Model, Surpasses DeepSeek V3
SenseTime Technology has officially launched its new 'Daily New' Fusion Large Model, which marks a significant leap forward in the field of artificial intelligence. This advanced model has shown remarkable capabilities in multimodal information processing and deep reasoning, securing the top position in two authoritative evaluation rankings.
The 'Daily New' model scored highly in the '2024 Annual Report on Chinese Large Model Benchmark Evaluation' by the SuperCLUE agency, sharing the top spot with DeepSeek V3 with a score of 68.3. The model also excelled in the multimodal evaluation conducted by OpenCompass, surpassing GPT-4o by a considerable margin.
One of the standout features of the 'Daily New' Fusion Large Model is its ability to integrate various types of information, including text, images, and videos. This advancement in native multimodal training opens up new possibilities for the industry. In the SuperCLUE evaluation, the model achieved top marks in both humanities and science, securing a global first place in humanities with a score of 81.8. It also led in the scientific category, ranking first in China with a score of 78.2 in computational dimensions.
The 'Daily New' model’s advanced capabilities allow it to handle complex information streams simultaneously. It can efficiently process ambiguous text, analyze data charts, and provide in-depth reasoning. These attributes make the model particularly effective in practical applications across diverse sectors, including autonomous driving, video interaction, office education, finance, and industrial manufacturing.
During its development, SenseTime utilized large datasets containing both text and images, enabling the creation of interactive connections between different modalities. This synthesis technology enhances the model’s understanding of multimodal information, making it more adaptable to various business scenarios. The company also developed a comprehensive set of cross-modal tasks, which provided a strong foundation for training the 'Daily New' Fusion Large Model. By utilizing innovative training methods, SenseTime’s model is poised to meet the diverse needs of users and facilitate a positive cycle of application implementation and model iteration.
Key Points
- SenseTime’s 'Daily New' Fusion Large Model demonstrates outstanding performance, ranking first in both humanities and science evaluations.
- The model is particularly adept at handling multimodal information, offering strong capabilities in areas such as autonomous driving, finance, and education.
- SenseTime's innovative training methods have set the foundation for continued success in various industry applications.
