Core Viewpoint - SenseTime's "SenseNova-V6.5 Pro" has achieved the highest score of 82.2 on the OpenCompass Multi-modal Academic Leaderboard, surpassing top international models like Gemini 2.5 Pro and GPT-5, marking it as one of the strongest multi-modal models globally [1][2][3] Group 1: Model Performance and Technology - "SenseNova-V6.5 Pro" is the latest achievement of SenseTime under its multi-modal general intelligence technology strategy, demonstrating significant advancements in multi-modal information perception and processing, which are essential for achieving AGI [1][3] - The model has successfully integrated "image-text interleaved thinking," allowing it to combine logical and visual thinking, thus enabling graphical representation of certain thought processes [3][4] - The model's reasoning performance has significantly improved through a new paradigm based on reinforcement learning, particularly in areas such as mathematics, coding, GUI operations, and high-level tasks [4] Group 2: Efficiency and Cost-Effectiveness - "SenseNova-V6.5 Pro" features an updated architecture with a lightweight visual encoder and a deepened MLLM backbone, achieving over three times efficiency improvement while maintaining performance, thus optimizing the performance-cost curve [4] - The model's cost-effectiveness is superior to that of international models like Gemini 2.5, indicating a strong competitive edge in the market [4] Group 3: Strategic Vision - SenseTime aims to build a leading general multi-modal model through a comprehensive strategy that integrates infrastructure, models, and applications, focusing on real-world scenarios to enhance end-to-end product technology competitiveness [4] - The company is committed to advancing multi-modal AI from the digital space into the physical world, providing end-to-end value in real scenarios [4] Group 4: Evaluation Framework - The OpenCompass evaluation system, launched by the Shanghai AI Laboratory, provides a comprehensive assessment platform for large models, covering various capabilities and specialized fields, and is regarded as an important reference for evaluating the application value of large models [5]