Recently, the globally recognized large model "arena" LMArena released its latest model ranking. According to the latest data obtained by AIbase, Baidu's new generation model ERNIE-5.0-Preview-1203 has officially登上 the text list with a high score of 1451. Notably, this achievement has enabled it to claim the top position among domestic large models, marking a milestone breakthrough for domestically developed native large models in international mainstream evaluation systems.

image.png

In actual evaluation performance, ERNIE-5.0 Preview demonstrated strong overall capabilities, especially excelling in creative writing and understanding high-complexity complex instructions. According to feedback from the LMArena ranking, the model is able to accurately capture users' deep needs, generating text content that is not only well-structured but also exceeds several domestic and international mainstream models in logical consistency. Previously, this model had already shown leading advantages in the visual understanding ranking, firmly holding the top position domestically.

As Baidu's flagship native multi-modal large model released in November of this year, ERNIE-5.0 has an astonishing parameter count of 2.4 trillion. AIbase learned that this model uses advanced native multi-modal unified modeling technology. This means it is not simply combining different models, but instead achieves native integration of text, images, audio, video, and other types of information. Whether in terms of understanding or generation capabilities, ERNIE-5.0 demonstrates the synergistic effects of multi-modal large models, further lowering the threshold for handling complex tasks.

Since its first release in 2019, the ERNIE series model has undergone multiple iterations. The impressive performance of the 5.0 Preview version on LMArena is not just about stacking technical parameters, but also reflects the maturity of domestic large models in handling Chinese contexts and complex logic tasks.

Key Points:

  • 🏆 New High Ranking: ERNIE-5.0 Preview climbed to the top of the LMArena text list with a score of 1451, ranking first domestically and second globally.

  • 🧠 Excellent Performance: The model performs exceptionally well in creative writing and complex task scenarios, demonstrating strong instruction following and logical generation capabilities.

  • 🌐 Ultra Large Scale: The model has a parameter count of 2.4 trillion and adopts native multi-modal modeling technology, supporting comprehensive interaction across text, images, audio, and video.