RM新时代有限公司

Recently,, CloudWalk Technology's CongRong Large Model has made significant progress inthe comprehensive evaluation authoritative platformOpenCompassofin the field of multimodal evaluation.have made significant progress.

The latest evaluation results show that the average score of CloudWalk Technology's CongRong Large Model in this system is65.5, ranking it among the top three globally, surpassing Google'sGemini-1.5-ProandGPT-4v, and second only toGPT-4o(69.9) andClaude3.5-Sonnet(67.9). In the domestic market, the CongRong Large Model's score also surpassedInternVL-Chat(61.7)andGLM-4V(60.8), ranking first.

Figure1: OpenCompassMultimodal Leaderboard

OpenCompassThe large model open evaluation system is a complete open-source and reproducible evaluation framework launched by Shanghai Artificial Intelligence Laboratory..In terms of OpenCompass multimodal evaluation,it adopts8arepresentativedatasets,objectively quantifying the capabilities of multimodal large models from multiple perspectives,with evaluation dimensions covering object detection, text recognition, action recognition, image understanding and relationship reasoning, art and design, business, science, health and medicine, humanities and social sciences, technology and engineering, mathematical reasoning, and other aspects.

 Figure2: CongRong Large Model-2.0Multimodal Capability Examples

In this evaluation, the CongRong Large Model performed excellently on6datasets, ranking first domestically (MMbench, MMStar, MathVista, HallusionBench, AI2D, OCRBench), especially on theOCRBenchtest setachieving the highest global827points (out of1000points), and higher than the second place byGLM-4v 13points, furtherenhancing the applicability of the CongRong Large Model in business scenarios such as text recognition, text-centric visual question answering, document-oriented visual question answering, and key information extraction.

Figure3: OpenCompassDemonstration of Chinese Large Model Multimodal Capabilities

The excellent performance of the CongRong Large Model in this systemrelies on CloudWalk Technology's self-developed efficient multimodal processing architecture andadvanced computing technology,, achieving efficient multimodal data processing capabilities, enabling efficient fusion and switching between vision and language tasks,andmaximizing the utilization of computing resources, ensuring high performance and response speed when processing large-scale multimodal data,makingthe model training process more efficient, with faster convergence and more stable performance.

It alsobenefits from CloudWalk Technology'slong-termaccumulation and continuous innovation in the fields ofvision and language.deep accumulation and continuous innovation in the field.

 Figure4: CongRong Large Model-2.0Multimodal Capability Examples

Previously, the CongRong Large Model hadset world records multiple times in the fields of vision and cross-modality,10set world records multiple times, and its comprehensive performance has been evaluated by third parties such asSuperClue, C-Eval, ranking among the top five globally..

As a platform enterprise focused onhuman-machine collaborationtechnology research and development,platformCloudWalk Technology has been actively promoting the development and application ofAIagent and large model technologies.development and application.

With the rapid development of artificial intelligence technology, multimodal large models have become the core engine driving industrial transformation. The outstanding performance of the CongRong Large Model in theOpenCompasslarge model open evaluation system, is not only a recognition of CloudWalk Technology's technological innovation strength but alsoaccumulation and continuous innovation in the fields ofsets a benchmark for the industry,, inspiring global technology companies to scale new heights in the new round of AI competition.