Baidu has officially released its new AI model Ernie 5.0, which processes text, images, audio, and video within a unified architecture. According to the LMArena ranking as of January 15, 2026, the model Ernie-5.0-0110 scored 1,460 points, securing 8th place globally and 1st place among all Chinese models.

The model ranks on par with OpenAI’s GPT-5.1 (High) and outperforms Google’s Gemini 2.5 Pro as well as Anthropic’s Claude Sonnet 4.5. The closest Chinese competitor is GLM-4.7 from Zhipu AI. In the mathematics category, Ernie 5.0 placed second worldwide, trailing only GPT-5.2 (High).

The LMArena ranking is determined from numerous anonymous paired comparisons, in which users choose the better model answer.
The LMArena ranking is determined from numerous anonymous paired comparisons, in which users choose the better model answer. 

LMArena rankings are calculated using large-scale anonymous pairwise comparisons, where users select the better response between two models.

Ernie 5.0 is built on a Mixture-of-Experts (MoE) architecture with approximately 2.4 trillion parameters, of which less than 3% are activated per query, significantly improving computational efficiency.

Currently, the model is available exclusively via ernie.baidu.com. Unlike previous releases, Baidu has not yet published the model weights, nor released a technical report or detailed documentation.

Most recently, Baidu introduced Ernie-4.5-VL-28B-A3B-Thinking, an open model capable of reasoning over images, including zooming and visual text analysis