Tencent’s Hunyuan AI Model Tops CLUE Leaderboard with Record Score

Tencent’s “Hunyuan” AI large model achieved a record‑breaking 80.888 points on the CLUE benchmark, securing first place and setting a new historical high for the classification leaderboard.

Earlier, the model also topped five major cross‑modal retrieval leaderboards, including MSR‑VTT, demonstrating its strong multimodal capabilities.

Developed jointly by the TEG Data Platform and Machine Learning Platform teams with open‑source collaboration from Tencent Advertising technology, Hunyuan is a massive pre‑training model that supports natural language understanding, computer vision, multimodal content comprehension, copy generation, and visual generation.

Leveraging the GPU power and training acceleration framework of Tencent’s Taiji machine‑learning platform, Hunyuan achieves stable and efficient model iteration.

Its training adopts a curriculum‑learning approach, progressing from simple to complex tasks, which improves performance, enhances generality and reusability, and reduces training costs.

The CLUE victory confirms Hunyuan’s leading position in natural language processing and signals its expanding capabilities across various business scenarios.

Currently, Hunyuan is deployed in WeChat Search, Tencent Ads and other services, boosting search experience and ad recommendation precision, with plans to empower more applications in the future.