SenseTime’s SenseNova Claims Top Spot in OpenCompass Global Multi-Modal Model Academic Leaderboard

Shanghai Artificial Intelligence Laboratory’s leading evaluation platform, OpenCompass, has released its latest Multi-modal Academic Leaderboard. Data shows that SenseTime’s SenseNova-V6.5 Pro multi-modal large model has secured the first position with an overall score of 82.2, outperforming leading international models such as Gemini 2.5 Pro and GPT-5.

This achievement establishes SenseTime’s SenseNova V6.5 as one of the world’s most powerful multi-modal large models, which is capable of understanding diverse data types including text, videos, images, and audio. It also exemplifies the success of SenseTime’s "multi-modal general intelligence" technological strategy. Lin Dahua, Co-Founder, Executive Director, and Chief Scientist of SenseTime, stated, "The core of intelligence lies in the ability to interact autonomously with the external world. The capability to perceive and process multi-modal information is a fundamental requirement for Artificial General Intelligence (AGI). It enables AI to receive and integrate information through multiple senses such as vision and hearing – just like humans – thereby achieving deeper understanding and reasoning, which is an indispensable path toward AGI."

SenseNova V6.5 represents SenseTime’s latest breakthrough under its multi-modal technology-driven approach. As a pioneer in China to develop the "text-image interleaved chain-of-thought" technology, SenseNova V6.5 can combine logical thinking and Imaginative thinking like humans, converting parts of the thought process into graphical expressions, thus possessing true multi-modal thinking capabilities, becoming the first commercial-grade large model in China to achieve imaginative thinking and interleaved text-image reasoning.

Furthermore, leveraging a new paradigm centered on chain-of-thought as a carrier and reinforcement learning as the primary method, SenseNova V6.5 can continuously enhance its reasoning capabilities through the algorithmic closed loop of "generation-verification-learning," significantly enhancing its reasoning performance, especially in dimensions such as mathematics, coding, GUI operations, chart analysis, and higher-order tasks.

Through architectural upgrades, SenseNova V6.5 adopts a lightweight visual encoder while deepening the MLLM (Multi-modal Large Language Model) backbone network, achieving over 3x the efficiency while maintaining the same performance, significantly optimizing the performance-cost curve, with cost-effectiveness outperforms international models like Gemini 2.5.

In the journey toward AGI through multi-modality, SenseTime is committed to building an industry-leading general multi-modal large model through a comprehensive strategy of “infrastructure-model-application,” while developing end-to-end product technology competitiveness driven by real-world scenarios. Based on the continuously evolving capabilities of multi-modal large models, SenseTime is dedicated to advancing multi-modal AI from the digital space into the real physical world, providing end-to-end value in real-world scenarios.

The OpenCompass large model open evaluation system is a one-stop, large model evaluation platform launched by the Shanghai Artificial Intelligence Laboratory, aimed at providing fair, open, and reproducible evaluation standards. It covers various aspects of general capabilities and specialized fields such as language, multi-modality, safety, embodied intelligence, finance, and healthcare, offering a comprehensive diagnostic professional leaderboard for the true capabilities of large models. The multi-modal model public academic leaderboard selects influential open-source academic evaluation sets from both domestic and international sources to evaluate industry multi-modal models, employing a combination of subjective and objective evaluation methods, utilizing strategies like CircularEval and LLM-as-a-Judge, and is regarded by the industry as an important reference reflecting the 'application value' of large models.

For the OpenCompass Multi-modal Model Public Academic Leaderboard, please visit: https://rank.opencompass.org.cn/leaderboard-multimodal/?m=REALTIME