Baidu Releases Wenxin Large Model 5.1: Search Capabilities Rank First in the Country, Pre-training Cost is Only 6% of Industry Standards

At a time when the competition in artificial intelligence is intensifying, domestic large models have once again delivered an impressive performance. On May 9, Baidu officially launched its new generation of foundational large model - ERNIE Bot 5.1. The model has now been fully launched on Baidu Qianfan Model Square and the ERNIE Bot official website, officially opening experience to enterprise users and developers.

The core highlight of this update lies in its high "cost-effectiveness" and powerful search integration capabilities. According to the official introduction, ERNIE Bot 5.1 adopts an innovative "Multi-dimensional Elastic Pre-training" technology. This technology first emerged when ERNIE Bot 5.0 was released. Its essence lies in achieving "one-time training, generating multiple scale models." As the latest stage result of this technology, ERNIE Bot 5.1 has significantly optimized the parameter scale on the basis of fully inheriting the knowledge reserve of version 5.0: the total parameters are compressed to about one-third, and the activated parameters are compressed to about half.

Notably, through this technology, Baidu has reduced the pre-training cost to about 6% of that of industry models of similar scale, while maintaining continuous leadership in basic performance. In the latest ranking of the international authoritative large model arena LMArena, ERNIE Bot 5.1 scored 1223 points, ranking first in the search list domestically and fourth globally, becoming the only domestic model among the top ranks on the list.

The so-called "search capability" is not simply the accumulation of information, but the model's ability to quickly retrieve, accurately integrate, and logically generate multi-source information. This means that ERNIE Bot 5.1 can provide answers with higher consistency and stronger reliability, showing great application potential in handling complex business scenarios, intelligent assistant interactions, and enterprise knowledge management.

In addition to its search expertise, ERNIE Bot 5.1 also performs strongly in other dimensions. Test data shows significant improvements in its Agent (intelligent agent) capabilities, surpassing DeepSeek-V4-Pro; in creative writing, its performance is comparable to Gemini 3.1 Pro; and in core reasoning capabilities, it is approaching the level of globally leading closed-source models.

Previously, the ERNIE series models have repeatedly ranked among the top domestic models in text and visual understanding lists. At the end of April this year, the preview version of ERNIE Bot 5.1 had already surpassed several mainstream domestic and international models on the LMArena text list.

It is reported that the Baidu AI Developer Conference will be held in Beijing from May 13 to 14. At that time, more technical details about ERNIE Bot 5.1 and specific plans for commercialization will be further disclosed. The industry is watching closely to see whether domestic large models can achieve a breakthrough by taking this high-performance path.

Baidu Releases Q1 2026 Earnings Report: Total Revenue of 32.1 Billion Yuan, AI Business Revenue Share Exceeds 52% for the First Time

Baidu's Q1 2026 report shows total revenue of 32.1 billion yuan, with general business revenue of 26 billion yuan, up 2% year-on-year. AI business revenue reached 13.6 billion yuan, surpassing 52% of general business for the first time, becoming a core growth engine. Executives discussed large model iteration, computing power layout, and commercialization paths during the earnings call.....

Baidu Establishes a Model Committee (BMC) to Coordinate BMU and AMU for Integrated R&D

On May 15, Baidu established the Baidu Model Committee (BMC) to oversee large model R&D and deployment, achieving a full-chain closure from technical exploration to business application. The committee integrates basic and applied model development departments, promoting 'technology-application integration,' marking a new efficient phase for Baidu's large model strategy.....

Li Yanhong: In the AI Era, Token Daily Active Users Are Not the New Benchmark; the Number of Intelligent Entities Is the New Standard

At the 2026 Baidu AI Developer Conference, Robin Li proposed shifting the key metric for AI industry prosperity from 'Token consumption' to 'Daily Active Agents' (DAA). He argued that Tokens merely reflect computational investment and cost, not output value; only by focusing on the number of results delivered by agents to humans can the essence of AI industry development be captured.....

Baidu Launches the Miaoda App Mobile: 99% of Code is Self-Generated, Early Developers Have Earned Millions

Baidu launched the no-code development platform Miaoda "App mobile at Create2026 conference, completed by a single developer, with 99% of the code generated automatically by the platform, showcasing breakthroughs in intelligent code engineering. The enterprise version was also launched simultaneously to promote the transition of AI from content generation to practical production.

Is AI Spread Rumors Illegal? China's First Case of Defamation Involving AI Large Models Concludes with a Final Judgment! The Court Ordered Baidu to Apologize

Recently, China's first case of defamation involving AI large models concluded with a final judgment in Nanjing. The case originated from lawyer Li Xiaoliang discovering that Baidu's AI smart answer generated false information, stating that he was sentenced to three years in prison, which severely damaged his reputation. The case highlights the legal risks of AI-generated content, and ultimately, the court ruled that the infringement was established.