Tencent's Hybrid Large Model Tops Domestic List in August's SuperCLUE-V Benchmark

TapTechNews August 5th news, according to today's report by Tencent Technology, the Chinese multimodal large model SuperCLUE-V benchmark August list was released, and Tencent's Hybrid Large Model ranked first among domestic large models (71.95 points).

Tencent Technology claims that this model accurately identifies image elements and generates natural language descriptions, comprehensively understanding and perceiving details. This evaluation covered 12 highly representative multimodal understanding large models at home and abroad. The Tencent Hybrid Model scored 71.95 in the multimodal basic ability and application ability, ranking second on the overall list, only next to GPT-4o. GPT-4o achieved 74.36 points, leading the multimodal benchmark, with both basic multimodal cognitive ability and application ability showing performances of over 70+ points, having certain leading advantages in technology and application.

SuperCLUE evaluation said that in terms of basic ability, there is still a certain gap between domestic large models and overseas models, especially in fine-grained visual cognitive tasks, there is a 5-point gap between the best models at home and abroad, and further optimization and improvement of multimodal deep cognitive ability are needed.

This evaluation selected 4 overseas models and 8 domestic representative multimodal models. Among them, in order to further assess the different progress of open source and closed source, this participating model includes 4 open source models and 8 closed source models.