360 Smart Brain Model: Leading the Pack in AI Innovation

On July 9, SuperCLUE, a leading Chinese model evaluation organization, released its "2024 Mid-Year Report on Chinese Large Model Benchmark Testing." In the June assessments, the 360 Zhinao large model (360GPT2-pro) achieved a score of 72, surpassing GPT-3.5-Turbo-0125 and ranking fourth among Chinese large models. It remains firmly within the top tier and has been classified as an "Outstanding Leader" in the SuperCLUE model quadrant.

The report evaluated 33 representative large models from both within China and globally, utilizing a comprehensive testing framework that reflects general capabilities across three dimensions: science, humanities, and technical tasks. The findings indicate a three-tier market structure for Chinese large models, with 360GPT2-pro leading the first tier. It excels in foundational technology and application scenarios, driving advancements in China's large model landscape and significantly narrowing the gap with international counterparts.

Notably, 360GPT2-pro scored an impressive 75 in the humanities section, ranking second overall. Its performance in generating creative content ranked first, positioning it at a similar level to GPT-4. Compared to other leading models, 360 Zhinao demonstrates superior competitiveness in language processing quality, content generation, and comprehension.

In the SuperCLUE Chinese Retrieval-Augmented Generation (RAG) assessment, 360 Zhinao RAG secured fourth place, further closing the gap with international models and showcasing its capability in Chinese retrieval generation. Previously, 360 announced the open sourcing of two models, 360Zhinao-search and 360Zhinao-1.8B-Reranking, both aimed at enhancing long-text retrieval performance. These models optimize the retrieval and ranking phases in RAG technology, complemented by the previously released 360K long text capabilities, providing an effective solution to reduce the likelihood of errors in large model outputs.

Currently, the 360 Zhinao large model has developed a diversified matrix of large models supporting various parameters and multimodal applications. It recently passed national compliance tests, placing it within the top tier in terms of overall capabilities. As 360 continues to strengthen its technical abilities, it accelerates the application of its models to create innovative solutions addressing diverse user needs. This year, 360 has revitalized its flagship products, launching the 360 AI Browser and 360 AI Search. The latter has consistently topped the global AI product growth charts, solidifying its lead in the Chinese AI search engine market.