Home
Baidu Launches Wenxin Large Model 5.1, Tops Search Benchmarks with Pre-training Cost at 6% of Industry Average
As competition in artificial intelligence intensifies, domestic large language models have once again delivered impressive results. On May 9, Baidu officially launched its next-generation foundation model, ERNIE Bot 5.1, now fully available on the Baidu Qianfan Model Square and the ERNIE Bot official website, opening access to enterprise users and developers.
The key highlights of this update are strong cost-efficiency and powerful search integration capabilities. According to Baidu, ERNIE Bot 5.1 adopts an innovative "Multi-dimensional Elastic Pre-training" technology, first introduced with ERNIE Bot 5.0. The core idea is to enable "one-time training, generating multiple scale models." As the latest milestone of this approach, ERNIE Bot 5.1 achieves significant parameter optimization while retaining all the knowledge from version 5.0: total parameters are compressed to roughly one-third, and activated parameters are reduced to about half.

Notably, this technology allows Baidu to cut pre-training costs to about 6% of comparable industry models while maintaining a leading edge in fundamental performance. In the latest ranking of the international authoritative large model arena LMArena, ERNIE Bot 5.1 scored 1223 points, ranking first among domestic models in search-related tasks and fourth globally—making it the only domestic model in the top tier of the list.
Search capability here goes beyond simple information retrieval. It refers to the model's ability to quickly fetch, accurately synthesize, and logically generate answers from multiple sources. This means ERNIE Bot 5.1 can deliver more consistent and reliable responses, showing strong potential across complex business scenarios, intelligent assistant interactions, and enterprise knowledge management.

In addition to search expertise, ERNIE Bot 5.1 excels in other areas. Test data reveal significant improvements in its Agent (intelligent agent) capabilities, surpassing DeepSeek-V4-Pro. In creative writing, its performance is on par with Gemini 3.1 Pro. And in core reasoning, it is approaching the level of globally leading closed-source models.
Previously, ERNIE series models consistently ranked among the top domestic models in text and visual understanding benchmarks. At the end of April this year, the preview version of ERNIE Bot 5.1 had already outperformed several major domestic and international models on the LMArena text leaderboard.
Baidu's AI Developer Conference is scheduled for May 13–14 in Beijing, where more technical details about ERNIE Bot 5.1 and specific commercialization plans will be revealed. The industry is watching closely to see if domestic large models can achieve a breakthrough by following this high-performance path.
Related article
DeepMind CEO Hassabis: I sleep six hours a day, usually feel energetic around 1 a.m.
Fortune recently featured an interview with Demis Hassabis, CEO of Google DeepMind, revealing his unconventional approach to rest and productivity. Hassabis disclosed that he sleeps very little, structuring his waking hours into two distinct work blo
OpenAI, Anthropic Vie for Market Share Despite Revenue Shortfalls
Despite recent reports suggesting OpenAI missed revenue targets, creating pressure on tech stocks this Tuesday, private AI lab investors remain resilient. Seasoned backers have confirmed they will not reduce investment despite negative media coverage
California AV Compliance: A New Era of Tickets, Geofences, and 1M Miles
Guident operates an AuveTech shuttle in South Florida, managing a four-mile route in West Palm Beach and a one-mile route in Boca Raton using its remote monitoring technology. | Credit: GuidentCalifornia is redefining the regulatory landscape for dri
Related Special Topic Recommendations
Comments (0)
0/500
As competition in artificial intelligence intensifies, domestic large language models have once again delivered impressive results. On May 9, Baidu officially launched its next-generation foundation model, ERNIE Bot 5.1, now fully available on the Baidu Qianfan Model Square and the ERNIE Bot official website, opening access to enterprise users and developers.
The key highlights of this update are strong cost-efficiency and powerful search integration capabilities. According to Baidu, ERNIE Bot 5.1 adopts an innovative "Multi-dimensional Elastic Pre-training" technology, first introduced with ERNIE Bot 5.0. The core idea is to enable "one-time training, generating multiple scale models." As the latest milestone of this approach, ERNIE Bot 5.1 achieves significant parameter optimization while retaining all the knowledge from version 5.0: total parameters are compressed to roughly one-third, and activated parameters are reduced to about half.

Notably, this technology allows Baidu to cut pre-training costs to about 6% of comparable industry models while maintaining a leading edge in fundamental performance. In the latest ranking of the international authoritative large model arena LMArena, ERNIE Bot 5.1 scored 1223 points, ranking first among domestic models in search-related tasks and fourth globally—making it the only domestic model in the top tier of the list.
Search capability here goes beyond simple information retrieval. It refers to the model's ability to quickly fetch, accurately synthesize, and logically generate answers from multiple sources. This means ERNIE Bot 5.1 can deliver more consistent and reliable responses, showing strong potential across complex business scenarios, intelligent assistant interactions, and enterprise knowledge management.

In addition to search expertise, ERNIE Bot 5.1 excels in other areas. Test data reveal significant improvements in its Agent (intelligent agent) capabilities, surpassing DeepSeek-V4-Pro. In creative writing, its performance is on par with Gemini 3.1 Pro. And in core reasoning, it is approaching the level of globally leading closed-source models.
Previously, ERNIE series models consistently ranked among the top domestic models in text and visual understanding benchmarks. At the end of April this year, the preview version of ERNIE Bot 5.1 had already outperformed several major domestic and international models on the LMArena text leaderboard.
Baidu's AI Developer Conference is scheduled for May 13–14 in Beijing, where more technical details about ERNIE Bot 5.1 and specific commercialization plans will be revealed. The industry is watching closely to see if domestic large models can achieve a breakthrough by following this high-performance path.
DeepMind CEO Hassabis: I sleep six hours a day, usually feel energetic around 1 a.m.
Fortune recently featured an interview with Demis Hassabis, CEO of Google DeepMind, revealing his unconventional approach to rest and productivity. Hassabis disclosed that he sleeps very little, structuring his waking hours into two distinct work blo
OpenAI, Anthropic Vie for Market Share Despite Revenue Shortfalls
Despite recent reports suggesting OpenAI missed revenue targets, creating pressure on tech stocks this Tuesday, private AI lab investors remain resilient. Seasoned backers have confirmed they will not reduce investment despite negative media coverage











