ERNIE-5.0 Tops LMArena Text Leaderboard as No. 1 Chinese Model
BEIJING — Baidu’s ERNIE-5.0-0110 model has claimed the top spot among Chinese large language models on the influential LMArena Text Arena leaderboard, scoring 1,460 points to rank No. 8 globally, the company announced.
LMArena released its latest rankings on January 15, 2026. The model, no longer labeled “Preview,” achieved the highest score of any Chinese entrant and became the only Chinese model to crack the platform’s global top 10. It now sits ahead of several leading international models, including OpenAI’s GPT-5.1-High and Google’s Gemini-2.5-Pro.
The result marks a significant milestone for Baidu’s ERNIE series, which has long been China’s flagship large language model family. According to Baidu’s official ERNIE Blog, the transition from preview to formal version and immediate ascent to the top of the Chinese cohort “demonstrates the model’s strong performance and reliability.”
Strong Performance Across Global Benchmarks
LMArena, formerly known as LMSYS Chatbot Arena, is a crowd-sourced, blind-comparison platform where users vote on model responses in real-world scenarios. The Text Arena leaderboard specifically evaluates pure language capabilities without multimodal components.
ERNIE-5.0-0110’s 1,460-point score places it eighth worldwide. Multiple reports confirm it surpassed OpenAI’s GPT-5.2, GPT-5.1-High, and other prominent Western models in the latest update. One independent analysis described the model as “a math wiz that beats OpenAI’s GPT 5.1,” highlighting particular strength in quantitative reasoning tasks.
Industry observers note that Baidu’s latest model is reportedly a 2.4 trillion parameter system, though Baidu has not publicly confirmed the exact parameter count in its official announcement. The model’s strong showing on a Western-hosted, user-voted benchmark is notable given the historical performance gap between leading Chinese and U.S. models on many public leaderboards.
Baidu’s Continued Push in Foundation Models
Baidu has invested heavily in developing indigenous large language models since launching the original ERNIE in 2019. The company positions ERNIE as both a general-purpose chatbot and a foundation model for enterprise applications across search, autonomous driving, and cloud services.
The removal of the “Preview” label on LMArena indicates that Baidu now considers ERNIE-5.0-0110 production-ready. This formal release follows a pattern seen with previous ERNIE iterations, where extended preview periods allowed the company to refine performance based on real user feedback before broader rollout.
Impact on Developers and the AI Industry
For Chinese developers and enterprises, the result strengthens confidence in domestically developed models for applications where data sovereignty, regulatory compliance, or integration with Baidu’s ecosystem are priorities. ERNIE models are accessible through Baidu’s cloud platform and have been integrated into the company’s search engine and other consumer products.
Globally, the ranking underscores the narrowing performance gap between leading Chinese and Western models on certain crowd-sourced benchmarks. While U.S. companies still dominate many academic and standardized evaluations, LMArena’s human-preference methodology appears to favor ERNIE-5.0’s current capabilities in real-world conversational and reasoning tasks.
What’s Next
Baidu has not yet disclosed a specific timeline for wider public release of ERNIE-5.0 beyond its current availability on LMArena and internal platforms. The company typically follows strong benchmark performances with expanded API access and integration into consumer-facing products such as its Ernie Bot chatbot.
Further details on model architecture, training data, context window size, and exact pricing for API access remain unavailable in the initial announcement. Industry watchers will be watching subsequent updates to see whether ERNIE-5.0 can maintain its position as newer models from OpenAI, Anthropic, Google, and other Chinese labs enter the arena.
LMArena rankings are updated regularly based on ongoing user votes, meaning positions can shift as more interactions are collected. Baidu’s current lead among Chinese models, however, represents its strongest showing to date on the prominent international platform.